Master of Science in Analytics
Blue Book for Bulldozers
- Predicting Bulldozer Sales Price
MSiA 420 Predictive Analytics
Emily Eunhee Ko, Yoojong Bang, Samuel Hillis
Benedict Lim, Joon Lim
0
Executive Summary .......................................................................................................... 2
Data Overview ................................................................................................................... 2
Detailed Model Analysis ................................................................................................... 4
Linear Regression ........................................................................................................ 7
Data Description .................................................................. Error! Bookmark not defined.
Results ................................................................................. Error! Bookmark not defined.
Ridge Regression ........................................................................................................ 7
Data Description ................................................................................................................. 8
Results ................................................................................................................................ 8
K-Nearest Neighbor Classification (KNN) ..................................................................... 8
Data Description .................................................................. Error! Bookmark not defined.
Results ................................................................................................................................ 8
Support Vector Machines Classification (SVM) ............................................................ 9
Data Description ................................................................................................................. 9
Results ...............................................................................................................................10
Naïve Bayes Classification..........................................................................................10
Data Description .................................................................. Error! Bookmark not defined.
Neural Networks .........................................................................................................10
Data Description .................................................................. Error! Bookmark not defined.
Results ................................................................................. Error! Bookmark not defined.
Boosting (GBM) ..........................................................................................................10
Data Description .................................................................. Error! Bookmark not defined.
Results ................................................................................. Error! Bookmark not defined.
Classification Trees (CART) ........................................................................................11
Data Description ................................................................................................................11
Results ................................................................................. Error! Bookmark not defined.
GAM ...........................................................................................................................12
Data Description .................................................................. Error! Bookmark not defined.
Results ................................................................................. Error! Bookmark not defined.
MARS .........................................................................................................................12
Data Description .................................................................. Error! Bookmark not defined.
Results ................................................................................. Error! Bookmark not defined.
Random Forest ...........................................................................................................12
Data Description ................................................................................................................13
Results ...............................................................................................................................14
PPR ............................................................................................................................14
Data Description .................................................................. Error! Bookmark not defined.
Results ................................................................................. Error! Bookmark not defined.
Conclusion ............................................................................. Error! Bookmark not defined.
1
Executive Summary
[]
Competition Overview
Competition Details
is a crowdsourcing platform for predictive modeling and analytics competitions
where companies post problems and data sets for researchers and analysts all over the world to
apply data mining techniques on. We entered into the “Blue Book for Bulldozers” competition
where the goal of the contest is to predict the sale price of a particular piece of heavy equipment
at an auction based on the equipment’s attributes such as year it was made and number of
machine hours on the equipment.
Forecasting Goal
The goal of the competition is to produce the most accurate forecasts of the equipment sale
price. The accuracy of the model is evaluated using the Root Mean Squared Log Error
(RMSLE), calculated using the formula:
𝑛
1
𝑅𝑀𝑆𝐿𝐸 = √ ∑(log(𝑌𝑖 + 1) − log(𝑌ℎ𝑎𝑡𝑖 + 1))2
𝑛
𝑖=1
Where Yi is the actual sale price of the ith piece of equipment, and Yhati is the predicted sale
price. Kaggle provides a test set on its website where participants can use their models to
forecast sale price, and submit their predictions online. Kaggle then calculates RMSLE based
on the above formula and maintains a leaderboard to rank the performance of the participant’s
models. Currently the top ranked model has a RMSLE of 0.2209. Here are the current top
models below:
Data Overview
There are initially 53 variables in this data set and it consists of one response variable, which is
sale price, and 52 predictors. Most of them are categorical variables, but there are some
continuous variables in the data set. Sale price, which is a response variable is a continuous
variable, and three out of 52 predictors are continuous variables as well:
MachineHourCurrentMeters, YearMade, and Tire_Size. Following is the whole data set and its
descriptions.
Table1. Data Description
2
Variable
Description
SalesID
MachineID
ModelID
datasource
unique identifier of a particular sale of a machine at auction
identifier for a particular machine; machines may have multiple sales
identifier for a unique machine model (i.e. fiModelDesc)
source of the sale record; some sources are more diligent about reporting
attributes of the machine than others. Note that a particular datasource may
report on multiple auctioneerIDs.
identifier of a particular auctioneer, i.e. company that sold the machine at
auction. Not the same as datasource.
year of manufacturer of the Machine
current usage of the machine in hours at time of sale (saledate); null or 0
means no hours have been reported for that sale
value (low, medium, high) calculated comparing this particular Machine-Sale
hours to average usage for the fiBaseModel; e.g. 'Low' means this machine
has less hours given it's lifespan relative to average of fiBaseModel.
time of sale
cost of sale in USD
Description of a unique machine model (see ModelID); concatenation of
fiBaseModel & fiSecondaryDesc & fiModelSeries & fiModelDescriptor
auctioneerID
YearMade
MachineHoursCurrentMeter
UsageBand
Saledate
Saleprice
fiModelDesc
fiBaseModel
fiSecondaryDesc
fiModelSeries
fiModelDescriptor
ProductSize
ProductClassDesc
State
ProductGroup
ProductGroupDesc
Drive_System
disaggregation of fiModelDesc
disaggregation of fiModelDesc
disaggregation of fiModelDesc
disaggregation of fiModelDesc
Don't know what this is
description of 2nd level hierarchical grouping (below ProductGroup) of
fiModelDesc
US State in which sale occurred
identifier for top-level hierarchical grouping of fiModelDesc
description of top-level hierarchical grouping of fiModelDesc
machine configuration; typcially describes whether 2 or 4 wheel drive
Enclosure
Forks
machine configuration - does machine have an enclosed cab or not
machine configuration - attachment used for lifting
Pad_Type
Ride_Control
Stick
machine configuration - type of treads a crawler machine uses
machine configuration - optional feature on loaders to make the ride smoother
machine configuration - type of control
Transmission
Turbocharged
Blade_Extension
Blade_Width
Enclosure_Type
Engine_Horsepower
Hydraulics
Pushblock
Ripper
Scarifier
Tip_control
Tire_Size
Coupler
Coupler_System
Grouser_Tracks
Hydraulics_Flow
machine configuration - describes type of transmission; typically automatic or
manual
machine configuration - engine naturally aspirated or turbocharged
machine configuration - extension of standard blade
machine configuration - width of blade
machine configuration - does machine have an enclosed cab or not
machine configuration - engine horsepower rating
machine configuration - type of hydraulics
machine configuration - option
machine configuration - implement attached to machine to till soil
machine configuration - implement attached to machine to condition soil
machine configuration - type of blade control
machine configuration - size of primary tires
machine configuration - type of implement interface
machine configuration - type of implement interface
machine configuration - describes ground contact interface
machine configuration - normal or high flow hydraulic system
Track_Type
Undercarriage_Pad_Width
Stick_Length
Thumb
machine configuration - type of treads a crawler machine uses
machine configuration - width of crawler treads
machine configuration - length of machine digging implement
machine configuration - attachment used for grabbing
3
Pattern_Changer
Grouser_Type
Backhoe_Mounting
Blade_Type
Travel_Controls
machine configuration - can adjust the operator control configuration to suit the
user
machine configuration - type of treads a crawler machine uses
machine configuration - optional interface used to add a backhoe attachment
machine configuration - describes type of blade
machine configuration - describes operator control configuration
Differential_Type
Steering_Controls
machine configuration - differential type, typically locking or standard
machine configuration - describes operator control configuration
Descriptive Statistics
Out of the initial 53 variables, we cut the data set down into 4 main predictor variables to predict
the numeric variable, SalePrice. We cut the data set down to a smaller data set mainly because
of computational issues. Filling up the sparse data set caused the size of the dataset to be
about 200mb, and R was not able to handle a dataset of this size even for a simple algorithm
such as linear regression.
Thus we used the Gradient Boosting Method (GBM) algorithm and identified which variables
were the most important variables in the data set and selected those variables as our main
variables for this study. In all, we identified four predictors (YearMade, State,
MachineHoursCurrentMeter, Enclosure, and fiProductClassDesc) and one response variable
(Sale Price) to run the following 12 models and will elaborate their descriptive statistics in this
section. Of the four predictors, machine hours and the year made are numeric variables, while
the rest are categorical variables.
Each algorithm used in this paper have specific limitations on what types of input it accepts.
Thus for each algorithm, a variation of this data set was used, and the variables selected are
briefly described in the “Model Analysis” section. Moreover, transformations on each of the
variables are performed as needed.
Sale Price (Response Variable)
Sale Price is ‘cost of sale (USD) in an auction.’ This is a numeric variable and following is
descriptive statistics for the variable.
Table2. Descriptive Statistics – Sale Price
Min.
4750
1st Qu.
14,500
Median
24,000
Mean
31,100
3rd Qu.
40,000
Max.
142,000
We initially saw the histogram of the response variable and observed its left-skewness. We,
therefore, took log 10 for this variable and found that degree of the skewness is reduced and the
histogram is more normalized. Therefore, we decided to take log 10 for our response variable.
4
Figure2. Histogram – Log10 Sale Price
Figure1. Histogram – Sale Price
MachineHoursCurrentMeter (Predictor)
MachineHoursCurrentMeter is ‘current usage of the machine in hours at time of sale (saledate)’;
null or 0 means no hours have been reported for that sale. This is a numeric variable and its
descriptive statistics is as follows.
Table3. Descriptive Statistics – MachineHoursCurrentMeter
Min.
0
1st Qu.
0
Median
0
Mean
3,458
3rd Qu.
3,025
Max.
248,300
NA’s
258,360
YearMade (Predictor)
YearMade means ‘year of manufacturer of the machine’ and it is another numeric variable in our
data set. Following is the descriptive statistics for YearMade.
Table4. Descriptive Statistics – YearMade
Min.
1919
1st Qu.
1988
Median
1996
Mean
1994
Enclosure (Predictor)
5
3rd Qu.
2001
Max.
2013
NA’s
38,185
Enclosure is one of the machine configurations and
a categorical variable which has five values under it.
According to data descriptions from Kaggle web
site, it means ‘if the machine has an enclosed cab
or not.’ There are five values under Enclosure and it
has two missing values.
State (Predictor)
Figure3. Histogram – Enclosure
State is a categorical variable and it means ‘the
state where sales occurred’. There are, in total, 53
values under ‘state’ including one ‘unspecified’
value which we considered missing value. The
number of missing value for ‘state’ variable is 2801.
Figure4. Histogram – State
6
fiProductClassDesc (Predictor)
Based on data description from Kaggle web site,
fiProductClassDesc means ‘description of 2nd level
hierarchical grouping (below ProductGroup) of
fiModelDesc’. It is a categorical variable and there
are in total 74 values and there is no missing value
under fiProductClassDesc.
Figure5. Histogram – fiProductClassDesc
Detailed Model Analysis
Linear Regression
Model Setup
Parameters
For the linear regression model, there are no parameters for us to perform a sensitivity analysis
on.
Data Description
The linear model allows for categorical variables, however in our dataset, the categorical
variables were relatively sparse, thus when we were trying to perform 10 fold cross validation,
there were cases where the training set did not contain certain categorical variables that the test
set did, causing the prediction to crash.
To take advantage of the available categorical data, we split the dataset by the product class,
and then performed a linear regression on a complete data set of SalePrice on the
standardized:
1) Number of Machine Hours on the Current Meter
2) Year made
Results
We ran this model for all 57 product classes, and the RMLSE for best model for each class
ranged from 0.220586345to 0.730789603 with an average RMLSE of 0.400408126. A detailed
breakdown of the RMLSE for each product class can be found in the appendix.
A closer look at the linear regression models created for each product class shows that the
average coefficient on the number of machine hours was -245.7014 while the average
coefficient on the year made was 11585.12. These coefficients make sense because it means
7
that the longer the machine has been used the lower the price, and the ‘younger’ the machine
the more valuable it is.
Ridge Regression
Model Setup
Parameters
For the ridge regression model, we ran cross validation on the linear regression model with 20
different lambda parameters ranging from 0.0004510930 to 1.
Data Description
The linear model allows for categorical variables, however in our dataset, the categorical
variables were relatively sparse, thus when we were trying to perform 10 fold cross validation,
there were cases where the training set did not contain certain categorical variables that the test
set did, causing the prediction to crash.
To take advantage of the available categorical data, we split the dataset by the product class,
and then performed a linear regression on a complete data set of SalePrice on the
standardized:
1) Number of Machine Hours on the Current Meter
2) Year made
Results
The results from the ridge regression were significantly worse than that from the simple linear
regression. We ran this model for all 57 product classes, and the RMLSE for best model for
each class ranged from 1.818759302 to 0.511227413 with an average RMLSE of 1.01693174.
A detailed breakdown of the RMLSE for each product class can be found in the appendix.
This result is could have stemmed from the fact that there is very little correlation between the
two predictor variables, and implementing ridge regression implemented an additional bias on
the coefficients and weakened the predictions. It is also important to note that the lambda that
had the lowest RMLSE for all the 57 product classes was the smallest lambda of 0.00045,
implying that the model would potentially have been better off without the additional constraint
on the size of the coefficients.
K-Nearest Neighbor Classification (KNN)
Model Setup
Parameters
For the KNN model, we experimented with a range of 3 to 10 nearest neighbors.
Data Description
The KNN model only allows for numerical variables, thus we were unable to include the other
categorical variables in the model. However, we split the dataset by the product class, and then
found the number of nearest neighbors that minimized RMLSE for each product class. The
variables used for each subset of data were:
1) Number of Machine Hours on the Current Meter
2) Year made
Results
8
We ran this model for all 57 product classes, and the RMLSE for best model for each class
ranged from 0.206888639 to 0.657082542 with an average RMLSE of 0.34215316. A detailed
breakdown of the RMLSE for each product class can be found in the appendix.
For most of the 57 product classes, the KNN algorithm picked 9 or 10 nearest neighbors as the
one that minimizes RMLSE.
Number of Nearest Neighbors, K
5
6
7
8
9
10
Number of Product Classes with
Associated Optimal K
1
1
4
6
12
33
Support Vector Machines Classification (SVM)
Model Setup
Parameters
A SVM model is a representation of the examples as points in space, mapped so that the
examples of the separate categories are divided by a clear gap that is as wide as possible. New
examples are then mapped into that same space and predicted to belong to a category based
on which side of the gap they fall on. In addition to performing linear classification, SVMs can
efficiently perform non-linear classification using a method, the kernel trick, to implicitly map
their inputs into high-dimensional feature spaces. This allows the SVM algorithm to fit the
maximum-margin hyper plane in a transformed feature space.
In this prediction model, other than the linear kernel, we focused on 3 types of kernels: the
sigmoid, polynomial, and the radial. Each kernel transforms the feature space in different nonlinear ways so as to be able to capture some of the non-linearity in the data set. The parameters
can be adjusted for each of the 3 kernels as follows:
•
•
•
•
Polynomial: Degree and Gamma
Sigmoid: Gamma and Coefficient
Radial: Gamma
Linear: Gamma
The range of gamma we experimented with ranges from 10-6 to 0.1, the degree ranges from 2 to
6, and the coefficients range from 0 to 3. We selected these parameters as these parameters
are commonly used by other researchers in the field. We tuned the model using these
parameters and for each combination of parameters we performed 10 fold cross validation to
get the average classification rate for each kernel.
Data Description
It is important to note that due to the size of the data, we were unable to run SVM on the entire
401,125 observations, thus we created subsets of the data based on the 57 product class
descriptions and individually did cross validation to find out the optimal parameters for each
product class. The variables used for each subset of data were:
1) State of sale
2) Type of enclosure
3) Number of Machine Hours on the Current Meter
4) Year made
9
Results
We ran this model for all 57 product classes, and the RMLSE for best model for each class
ranged from 0.208401176 to 0.519523311 with an average RMLSE of 0.329685653. A detailed
breakdown of the RMLSE for each product class can be found in the appendix.
Neural Networks
Model Setup
Parameters
Data Description
Results
Boosting (GBM)
Model Setup
Parameters
The parameters that we chose to vary for cross-validation purposes were interaction depth and
shrinkage. An interaction depth of 1 implies that the model is simply additive in nature while a
depth of “k” implies that there may be interaction present between combinations of k variables.
We chose to explore interaction depths up to 4. The shrinkage parameter is essentially the
lambda value that is used when developing trees for the model. We tested shrinkage values
{0.1, 0.2, … , 1.0}. For each model we also chose to fit 100 trees; we determined this was a
sufficient value heuristically by testing the prediction power of several models with 10, 100, and
1000 trees. While 1000 trees performed slightly better, the menial improvement did not justify
the huge increase in computational time due to the vastness of our dataset.
Data Description
The generalized boosting model is able to handle both quantitative and categorical data.
Additionally, the “gbm” function in R allows for datasets that contain null values. This was
especially useful for our dataset as many of the predictors were very sparse. We began with 52
predictor variables and reduced the set to 40 by removing meaningless and identical variables.
From these 40 variables, several initial boosting models were calculated to determine which
variables were useful in the model and chose to retain all variables with a relative influence
value greater than 0.1; this amounted to 24 predictors.
Results
For the 30 different combinations of shrinkage
parameters and interactions depths, we ran a 10-fold
cross validation to estimate the RMSLE of each pairing
of parameters. The maximum RMSLE came from the
model with a shrinkage parameter of 0.1 and interaction
depth of 2 (0.1597) while the smallest came from the
model with a shrinkage parameter of 0.6 and interaction
depth of 4 (0.1447). The average value among all
combinations was 0.1483.
Even still we see that the relative influence of several
predictors (in order: product class, year made, and
10
enclosure type) dominate the remaining 21 variables in relative influence in the model.
Classification Trees (CART)
Data Description
Since Regression Tree is capable of handling many categorical variables with lots of categories,
at least more than 100, and variables with lots of missing values that is more than 75% is null
for some variables, we decided to use all 52 predictor variables and prune the tree afterwards to
manage the over fitting issue.
Model Setup
In R, “rpart” package uses 𝑪𝒑 as a
method to penalize the tree to prevent
over fitting. As we can see the plot left,
the relative error converges to 0.43 after
𝑪𝒑 = 𝟎. 𝟏. Therefore, we chose our
shrinkage parameter, 𝑪𝒑, of 0.1 to prune
the tree. The corresponding size of tree
when 𝑪𝒑 = 𝟎. 𝟏 is the tree with 8 nodes.
As we can see on our final mode of
Regression Tree graph on the right, the
variable, “fiProductClassDesc ” is the most
influential variable to determine the sales price of
bulldozer. This is a categorical variable and the tree
divided into two parts based on whether the final
product class description is
“abcdefiprtvwBCDEFGHIJOQSZZ” or not. From the
plot, we can clearly notice that fiProductClassDesc ,
YearMade and Enclosure are significant predictor
variables to estimate Sales Price.
11
Results
With this Regression Tree Model, we ran
Repeated 10 Fold Cross Validation. This was
pretty manageable because each 10 fold cross
validation only took 115.892 seconds, that is
slightly less than 2 minutes. Interesting point is
each 10 fold cross validation, the RMLSE were not
vary much. The residual plot also shows errors
are randomly distributed.
From Regression Tree, we have the repeated 10
fold Cross Validated RMLSE of 0.3318572.
GAM
Model Setup
Data Description
The Generalized Addictive model only allows for numerical variables just like KNN model,
therefore we could not include categorical variables in our model. However, we split the dataset
by the product class, and then found the RMLSE for each product class.
The variables used for each subset of data were:
1) Number of Machine Hours on the Current Meter
2) Year made
Results
We ran the GAM for all 57 product classes, and we found that Generalized Additive Model did
predict better than Linear Regression and Ridge Regression with our data but was not so
different from K Nearest Neighbor. 10 Fold Cross Validation RMSLE from GAM ranged from
0.214957079 to 0.683211532. On average, GAM has 10 Fold CV RMSLE of 0.35196.
MARS
Model Setup
Multivariate adaptive regression spline (MARS) models use additive local linear regression
models that can handle both numerical and categorical data. The modeling technique creates
basis functions for each variable called “hinge functions”. These functions are essentially
piecewise linear functions; the value at which the piecewise function is divided is referred to as
the “knot”. Like most other non-parametric models, they seek to minimize the sum of squared
errors and are derived iteratively. The predicted value is a weighted sum of these hinge
functions. The algorithm consists of a forward and backward pass; these steps are very similar
to CART models in that the forward pass is meant to overfit the data while the backward pass is
12
meant to prune the least effective hinge functions. MARS software contains a built-in cross
validation technique that is used to determine when the backward pass should stop.
Data Description
Unfortunately the R implementation of multivariate adaptive regression splines cannot handle
null values. For this reason, we subset our predictor variables to the five least sparse: state,
enclosure type, machine hours, year made, and product class. Of the 400,000 observation only
about 120,000 had non-null values for each of these variables. The only parameter that we
looked at modifying for the MARS model was the number of interaction terms. Much like other
models, this value is the maximum number of variable interaction terms that the algorithm will
consider; values 1 through 4 were tested using 10-fold cross validation.
Results
For the interaction degree levels of 1 to 4, we ran a 10-fold cross validation to estimate the
RMSLE of each pairing of parameters. The maximum RMSLE came from the model with an
interaction depth of 1 (0.1512) while identical RMSLE values were found for models with
interaction terms of 2, 3 and 4 (0.1497). The average value among all combinations was
0.1501.
Random Forest
Model Setup
Recursive partitioning methods are one of the most popular and widely used supervised
learning methods for nonparametric regression and classification. Theoretically, random forest
is very powerful since it can handle large numbers of predictor variables even in the presence of
complex interactions. To use Random Forest in R, we have two packages, which are
“randomForest”, and “party.” They have slightly different characteristics and different abilities to
handle our data. Random forest in party package introduced the unbiased tree algorithm for
conditional inference trees and that enables learning unbiased forests. It is important to notice
that randomForest package and party package are different in the variable importance measure.
randomForest uses Gini importance, which is based on the Gini gain criterion employed in most
traditional classification algorithms while party uses conditional importance. Generally speaking,
the party package’s random forest is treated as an updated version of traditional random forest
algorithm offered by randfomForest package.
Data Description
For randomForest package, it does not take any null values and categorical variable has to
have less than 8 categories. But it offers the parallel computing technology with ‘foreach’
package, which save the time for running the algorithm significantly. Party’s random forest can
handle missing values and many categories but takes a long time to run the algorithm.
However, random forest offered from both packages required a large memory and high
computing power. We tested both random forest models with Mac Book Pro Retina 2012 model.
Since we have the considerably large sample size of 401,125 observations, this laptop cannot
handle more than 4 variables due to RAM and running time issues. (Even the random forest
benchmark dataset given by Kaggle consists of only one predictor variable.) Therefore, we
chose 4 variables based on CART Tree and GBM variable importance outcome and the
existence of null values.
1) MachineID
2) ProductGroup
13
3) YearMade
4) Saledate
As we can notice from the graph on the left,
ProductGroup is the most important
variable followed by YearMade and
MachineID. Salesdate is the least significant
to predict Sales Price among our 4
variables.
Results
We ran the 10 fold cross validation on
random forest for both packages.
randomForest package gave RMSLE of
0.4819 while party package gave
0.4889.
PPR
Model Setup
Parameters
Data Description
Results
Conclusion
The Kaggle competition is a fierce competition with sophisticated data scientists and analysts
across the world using cutting edge algorithms and techniques to achieve the best score. In our
project we attempted to apply as many of the algorithms we learned in class (and more) to get
comfortable with the dataset and understand the intricacies of the problem. One major issue we
faced was the size and how sparse the data was: R was unable to run the many algorithms
without running out of memory. Moreover, some algorithms were limited by whether it could take
in categorical, numeric data or both.
Therefore, for the purposes of this project, we ran three types of analysis for different types of
algorithms:
1) Prediction using subset of variables on all 400,000 observations
2) Prediction using all 4 variables splitting on one of the categorical data into 57 subsets
14
3) Prediction using 2 continuous variables splitting on one of the categorical data into 57
subsets
This approach allowed us to explore as many predictive analytic techniques as possible to give
us a sense of what algorithms would be most effective going forward, as well as what approach
we could take after taking into account data and computational limitations. We summarize our
findings as follow:
Key results:
1) GBM and MARS performed the best (algorithms using the 1st type of analysis)
 RMLSE of 0.1447 and 0.1497 respectively
2) We are able to identify which algorithms work best for each of the 57 subsets
 Given new observations, we can identify which product class it belongs to, and use the
vzalgorithm with the lowest RMLSE
 The best algorithm for each product class can be found in the appendix
In conclusion, we have taken a deep dive into exploring the potential predictive methods for the
Kaggle dataset. We find that even after taking out many of the sparse categorical variables, we
are still able to get good results after cross-validation within our training set. Future steps would
include [Joon to add in some sentences]
15
Appendix
Algorithm
Name
SVM
Product Class
Best Model
Ridge
KNN
Regression
Number of
Neighbours Best Lambda
Wheel Loader - Radial Model
110.0 to 120.0 Parameters:
Horsepower
0.1
Polynomial
Wheel Loader - Model
150.0 to 175.0 Parameters:
Horsepower
6&2
Skid Steer
Loader - 1351.0 Polynomial
to 1601.0 Lb
Model
Operating
Parameters:
Capacity
2&1
Hydraulic
Excavator,
Polynomial
Track - 12.0 to Model
14.0 Metric
Parameters:
Tons
3&2
Skid Steer
Loader - 1601.0 Polynomial
to 1751.0 Lb
Model
Operating
Parameters:
Capacity
4&3
Backhoe Loader
- 14.0 to 15.0 Ft
Standard
Digging Depth
Hydraulic
Excavator,
Track - 21.0 to
24.0 Metric
Tons
Hydraulic
Excavator,
Track - 3.0 to
4.0 Metric Tons
Radial Model
Parameters:
0.1
Radial Model
Parameters:
0.1
Radial Model
Parameters:
0.1
Polynomial
Wheel Loader - Model
350.0 to 500.0 Parameters:
Horsepower
5&1
Track Type
Tractor, Dozer - Radial Model
20.0 to 75.0
Parameters:
Horsepower
0.1
Hydraulic
Excavator,
Polynomial
Track - 19.0 to Model
21.0 Metric
Parameters:
Tons
3&2
Hydraulic
Excavator,
Radial Model
Track - 4.0 to
Parameters:
5.0 Metric Tons 0.1
Hydraulic
Excavator,
Radial Model
Track - 2.0 to
Parameters:
3.0 Metric Tons 0.1
Hydraulic
Excavator,
Track - 24.0 to Radial Model
28.0 Metric
Parameters:
Tons
0.1
Radial Model
Parameters:
0.1
Polynomial
Wheel Loader - Model
200.0 to 225.0 Parameters:
Horsepower
3&3
Hydraulic
Excavator,
Polynomial
Track - 50.0 to Model
66.0 Metric
Parameters:
Tons
4&1
Ridge
Neural
Regression GAM Networks PPR
SVM
Ridge
Best
Best number
Regression GAM parameters of terms
RMLSE
Linear
Regression
Ridge
Regression
GAM
Neural
Networks
PPR
RMLSE
RMLSE
RMLSE
RMLSE
RMLSE
RMLSE
8
Linear
Ridge
0.00045109 Regression Regression GAM 0.1&3
1
0.33575332
0.35076334
0.40837499
0.70163136
0.36868185
0.35538048
9
Linear
Ridge
0.00045109 Regression Regression GAM 1&4
2
0.31033801
0.33113282
0.35216205
0.67773369
0.3205651
0.35467294
8
Linear
Ridge
0.00045109 Regression Regression GAM 1&4
2
0.24630725
0.24410265
0.26810819
1.42895732
0.25893171
0.25797712
0.25001887 KNN: 8
0.24410265
10
Linear
Ridge
0.00045109 Regression Regression GAM 1&4
8
0.31138785
0.30577523
0.35633651
0.66627939
0.32760268
0.34002275
0.30577523
10
Linear
Ridge
0.00045109 Regression Regression GAM 1&4
2
0.23584601
0.24564255
0.26338521
0.9930175
0.25151523
0.25150977
0.31992246 KNN: 10
SVM:
Polynomial
Model
Parameters:
0.23786264 4&3
10
Linear
Ridge
0.00045109 Regression Regression GAM 1&5
0.24944523
SVM: Radial
Model
Parameters:
0.22956573 0.1
8
0.22810239
0.24000226
0.25716934
1.11142673
0.24268334
10
Linear
Ridge
0.00045109 Regression Regression GAM 1&5
1
0.34091423
0.35231256
0.39636337
1.05285981
0.36320212
0.38496036
10
Linear
Ridge
0.00045109 Regression Regression GAM 1&4
1
0.24431454
0.2512586
0.30901526
1.26680323
0.26348923
0.26263735
SVM: Radial
Model
Parameters:
0.413769 0.1
SVM: Radial
Model
Parameters:
0.24972948 0.1
Neural
Networks:
0.413769 1&5
SVM: Radial
Model
Parameters:
0.31090085 0.1
9
Linear
Ridge
0.00045109 Regression Regression GAM 1&5
4
0.43915666
0.44425007
0.51024605
0.79955844
0.45847529
0.48610911
10
Linear
Ridge
0.00045109 Regression Regression GAM 1&4
4
0.29899515
0.30327425
0.422715
0.84468004
0.3191405
0.33593979
10
Linear
Ridge
0.00045109 Regression Regression GAM 0.1&5
3
0.31277874
0.30947516
0.39973948
0.8394843
0.31990292
0.33802762
9
Linear
Ridge
0.00045109 Regression Regression GAM 0.1&4
1
0.26638053
0.26948051
0.3227393
1.06964016
0.28255372
0.29636534
0.32135175 KNN: 10
SVM: Radial
Model
Parameters:
0.28489108 0.1
Neural
Networks:
0.23215892 1&4
Linear
Ridge
0.00045109 Regression Regression GAM 1&4
8
0.23829788
0.2381777
0.2883876
1.24691859
0.24733567
0.24115873
7
Linear
Ridge
0.00045109 Regression Regression GAM 0.1&5
4
0.40188474
0.38703131
0.63985156
0.51122741
0.40418251
0.42743641
10
Linear
Ridge
0.00045109 Regression Regression GAM 0.1&4
9
0.41195991
0.46286548
0.49397032
0.96964403
0.46706404
0.4677637
0.413769 KNN: 7
SVM: Radial
Model
Parameters:
0.413769 0.1
10
Linear
Ridge
0.00045109 Regression Regression GAM 1&5
1
0.4383524
0.4294455
0.46345463
1.0676783
0.4363468
0.44888333
Neural
Networks:
0.413769 1&5
10
Linear
Ridge
0.00045109 Regression Regression GAM 1&3
2
0.51952331
0.52165395
0.63830237
0.80927636
0.53612937
0.56845338
Neural
Networks:
0.413769 1&3
Algorithm Name
SVM
KNN
Ridge
Regres
sion
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Neural
Networ
ks
Product Class
Best
Model
Number
of
Neighbo
urs
Best
Lambd
a
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Best
parame
ters
Wheel Loader 110.0 to 120.0
Horsepower
Radial
Model
Param
eters:
0.1
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
8
KNN
Best Model RMLSE
SVM: Radial
Model
Parameters:
0.35834164 0.1
0.33575332
SVM:
Polynomial
Model
Parameters:
0.31114744 6&2
0.31033801
10
Motorgrader 45.0 to 130.0
Horsepower
Linear
Regression
Linear
Regression
0.1&3
PPR
Best
num
ber
of
term
s
1
0.23584601
0.22810239
0.34091423
0.24431454
0.413769
0.29899515
0.30947516
0.26638053
0.23215892
0.38703131
0.41195991
0.413769
0.413769
SVM
KNN
Linear
Regress
ion
Ridge
Regress
ion
GAM
Neural
Networks
PPR
RMLSE
RMLSE
RMLSE
RMLSE
RMLSE
RMLSE
RML
SE
Best Model
RMLSE
0.33575
332
0.35076
334
0.40837
499
0.70163
136
0.36868
185
0.35
8341
64
SVM: Radial
Model
Parameters:
0.1
0.33575332
16
0.355380
48
Wheel Loader 150.0 to 175.0
Horsepower
Polyn
omial
Model
Param
eters:
6&2
9
Skid Steer
Loader - 1351.0
to 1601.0 Lb
Operating
Capacity
Polyn
omial
Model
Param
eters:
2&1
Hydraulic
Excavator, Track
- 12.0 to 14.0
Metric Tons
Polyn
omial
Model
Param
eters:
3&2
Skid Steer
Loader - 1601.0
to 1751.0 Lb
Operating
Capacity
Polyn
omial
Model
Param
eters:
4&3
Backhoe Loader
- 14.0 to 15.0 Ft
Standard
Digging Depth
Radial
Model
Param
eters:
0.1
Hydraulic
Excavator, Track
- 21.0 to 24.0
Metric Tons
Radial
Model
Param
eters:
0.1
Hydraulic
Excavator, Track
- 3.0 to 4.0
Metric Tons
Radial
Model
Param
eters:
0.1
Wheel Loader 350.0 to 500.0
Horsepower
Polyn
omial
Model
Param
eters:
5&1
Track Type
Tractor, Dozer 20.0 to 75.0
Horsepower
0.354672
94
0.31
1147
44
SVM:
Polynomial
Model
Parameters:
6&2
0.31033801
0.25893
171
0.257977
12
0.25
0018
87
KNN: 8
0.24410265
0.32760
268
0.340022
75
0.31
9922
46
KNN: 10
0.30577523
SVM:
Polynomial
Model
Parameters:
4&3
0.23584601
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
1&4
2
0.31033
801
0.33113
282
0.35216
205
0.67773
369
0.32056
51
8
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
1&4
2
0.24630
725
0.24410
265
0.26810
819
1.42895
732
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
1&4
8
0.31138
785
0.30577
523
0.35633
651
0.66627
939
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
9
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Radial
Model
Param
eters:
0.1
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Hydraulic
Excavator, Track
- 19.0 to 21.0
Metric Tons
Polyn
omial
Model
Param
eters:
3&2
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Hydraulic
Excavator, Track
- 4.0 to 5.0
Metric Tons
Radial
Model
Param
eters:
0.1
9
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Hydraulic
Excavator, Track
- 2.0 to 3.0
Metric Tons
Radial
Model
Param
eters:
0.1
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
2
0.23584
601
0.24564
255
0.26338
521
0.99301
75
0.25151
523
0.251509
77
0.23
7862
64
8
0.22810
239
0.24000
226
0.25716
934
1.11142
673
0.24268
334
0.249445
23
0.22
9565
73
SVM: Radial
Model
Parameters:
0.1
0.22810239
1
0.34091
423
0.35231
256
0.39636
337
1.05285
981
0.36320
212
0.384960
36
0.41
3769
SVM: Radial
Model
Parameters:
0.1
0.34091423
1
0.24431
454
0.25125
86
0.30901
526
1.26680
323
0.26348
923
0.262637
35
0.24
9729
48
SVM: Radial
Model
Parameters:
0.1
0.24431454
4
0.43915
666
0.44425
007
0.51024
605
0.79955
844
0.45847
529
0.486109
11
0.41
3769
Neural
Networks:
1&5
1&4
4
0.29899
515
0.30327
425
0.42271
5
0.84468
004
0.31914
05
0.335939
79
0.31
0900
85
SVM: Radial
Model
Parameters:
0.1
0.29899515
0.1&5
3
0.31277
874
0.30947
516
0.39973
948
0.83948
43
0.31990
292
0.338027
62
0.32
1351
75
KNN: 10
0.30947516
0.1&4
1
0.26638
053
0.26948
051
0.32273
93
1.06964
016
0.28255
372
0.296365
34
0.28
4891
08
SVM: Radial
Model
Parameters:
0.1
0.26638053
1&4
8
0.23829
788
0.23817
77
0.28838
76
1.24691
859
0.24733
567
0.241158
73
0.23
2158
92
Neural
Networks:
1&4
0.23215892
1&4
1&5
1&5
1&4
1&5
17
0.413769
Hydraulic
Excavator, Track
- 24.0 to 28.0
Metric Tons
Radial
Model
Param
eters:
0.1
Motorgrader 45.0 to 130.0
Horsepower
7
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Radial
Model
Param
eters:
0.1
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Wheel Loader 200.0 to 225.0
Horsepower
Polyn
omial
Model
Param
eters:
3&3
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Hydraulic
Excavator, Track
- 50.0 to 66.0
Metric Tons
Polyn
omial
Model
Param
eters:
4&1
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Hydraulic
Excavator, Track
- 40.0 to 50.0
Metric Tons
Polyn
omial
Model
Param
eters:
5&3
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
Hydraulic
Excavator, Track
- 33.0 to 40.0
Metric Tons
Radial
Model
Param
eters:
0.1
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
Skid Steer
Loader - 2201.0
to 2701.0 Lb
Operating
Capacity
Polyn
omial
Model
Param
eters:
6&2
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Wheel Loader 120.0 to 135.0
Horsepower
Radial
Model
Param
eters:
0.1
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Track Type
Tractor, Dozer 130.0 to 160.0
Horsepower
Radial
Model
Param
eters:
0.1
9
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Wheel Loader 275.0 to 350.0
Horsepower
Polyn
omial
Model
Param
eters:
5&1
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
Motorgrader 145.0 to 170.0
Horsepower
Radial
Model
Param
eters:
0.01
8
0.0004
5109
Linear
Regres
sion
Hydraulic
Excavator, Track
- 6.0 to 8.0
Metric Tons
Radial
Model
Param
eters:
0.1
10
0.0004
5109
Wheel Loader 60.0 to 80.0
Horsepower
Radial
Model
Param
eters:
0.1
10
0.0004
5109
10
10
GA
M
GA
M
4
0.40188
474
0.38703
131
0.63985
156
0.51122
741
0.40418
251
0.427436
41
0.41
3769
KNN: 7
0.38703131
0.1&4
9
0.41195
991
0.46286
548
0.49397
032
0.96964
403
0.46706
404
0.467763
7
0.41
3769
SVM: Radial
Model
Parameters:
0.1
0.41195991
1&5
1
0.43835
24
0.42944
55
0.46345
463
1.06767
83
0.43634
68
0.448883
33
0.41
3769
Neural
Networks:
1&5
0.413769
2
0.51952
331
0.52165
395
0.63830
237
0.80927
636
0.53612
937
0.568453
38
0.41
3769
Neural
Networks:
1&3
0.413769
0.417335
02
0.39
9107
92
SVM:
Polynomial
Model
Parameters:
5&3
0.38933384
0.378307
11
0.41
3769
SVM: Radial
Model
Parameters:
0.1
0.35468536
SVM:
Polynomial
Model
Parameters:
6&2
0.24643054
0.1&5
1&3
1&4
0.1&4
2
1
0.38933
384
0.35468
536
0.40959
054
0.37966
446
0.47060
631
0.37935
221
0.64644
183
0.93317
366
0.39932
584
0.36268
093
10
0.24643
054
0.24794
252
0.26862
243
1.02147
966
0.25447
553
0.261878
04
0.25
1501
75
3
0.28866
407
0.31282
209
0.35884
263
1.20026
808
0.31628
543
0.323098
69
0.30
0270
53
SVM: Radial
Model
Parameters:
0.1
0.28866407
1&3
2
0.36727
663
0.37044
194
0.55617
408
0.67896
448
0.37746
282
0.419319
67
0.36
7419
66
SVM: Radial
Model
Parameters:
0.1
0.36727663
GA
M
1&5
1
0.38746
149
0.38224
806
0.41222
417
0.90850
292
0.38308
271
0.408133
47
0.38
7442
19
KNN: 10
0.38224806
Ridge
Regres
sion
GA
M
1&4
1
0.49452
423
0.49534
362
0.52594
288
1.06452
298
0.49733
341
0.511988
56
0.41
3769
Neural
Networks:
1&4
Linear
Regres
sion
Ridge
Regres
sion
GA
M
1&5
2
0.29628
042
0.29065
79
0.33804
042
1.66925
021
0.30807
069
0.327035
04
0.30
8414
97
KNN: 10
Linear
Regres
sion
Ridge
Regres
sion
GA
M
10
0.24360
989
0.30076
637
0.38081
833
1.35694
763
0.31691
218
0.283193
04
0.25
2146
72
SVM: Radial
Model
Parameters:
0.1
0.1&4
1&4
1&3
18
0.413769
0.2906579
0.24360989
Other
Radial
Model
Param
eters:
0.1
Hydraulic
Excavator, Track
- 8.0 to 11.0
Metric Tons
Radial
Model
Param
eters:
0.1
Skid Steer
Loader - 1751.0
to 2201.0 Lb
Operating
Capacity
Polyn
omial
Model
Param
eters:
4&1
Motorgrader 170.0 to 200.0
Horsepower
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Radial
Model
Param
eters:
0.1
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Skid Steer
Loader - 1251.0
to 1351.0 Lb
Operating
Capacity
Polyn
omial
Model
Param
eters:
6&3
9
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Hydraulic
Excavator, Track
- 16.0 to 19.0
Metric Tons
Radial
Model
Param
eters:
0.1
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Hydraulic
Excavator, Track
- 0.0 to 2.0
Metric Tons
Polyn
omial
Model
Param
eters:
3&3
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
Backhoe Loader
- 15.0 to 16.0 Ft
Standard
Digging Depth
Polyn
omial
Model
Param
eters:
3&1
8
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Motorgrader 130.0 to 145.0
Horsepower
Radial
Model
Param
eters:
0.1
9
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Track Type
Tractor, Dozer 75.0 to 85.0
Horsepower
Radial
Model
Param
eters:
0.1
7
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Hydraulic
Excavator, Track
- 14.0 to 16.0
Metric Tons
Polyn
omial
Model
Param
eters:
3&3
9
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Track Type
Tractor, Dozer 85.0 to 105.0
Horsepower
Radial
Model
Param
eters:
0.1
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
5
10
10
GA
M
GA
M
GA
M
1&2
1&2
2
1
0.51496
064
0.29261
768
0.65708
254
0.31282
336
0.73078
96
0.35837
578
1.07683
14
1.24440
742
0.68321
153
0.32781
977
0.622165
83
0.41
3769
Neural
Networks:
1&2
0.334950
91
0.32
4035
19
SVM: Radial
Model
Parameters:
0.1
0.29261768
SVM:
Polynomial
Model
Parameters:
4&1
0.23070482
0.2609167
0.413769
6
0.23070
482
0.23247
183
0.25180
718
1.58901
583
0.24569
646
0.251898
18
0.23
3320
6
0.1&4
1
0.26091
67
0.27508
384
0.30386
9
1.59788
115
0.26710
015
0.274466
31
0.41
3769
SVM: Radial
Model
Parameters:
0.1
1&4
8
0.22193
511
0.21051
592
0.24853
04
0.74400
705
0.22992
763
0.231248
19
0.22
7690
3
KNN: 9
0.21051592
2
0.38197
897
0.39281
777
0.54958
854
0.65838
095
0.43397
318
0.457178
33
0.41
3769
SVM: Radial
Model
Parameters:
0.1
0.38197897
0.270345
13
0.25
7932
99
SVM:
Polynomial
Model
Parameters:
3&3
0.22410146
SVM:
Polynomial
Model
Parameters:
3&1
0.26441843
1&2
1&5
1&1
1&5
0.1&5
1&5
0.1&4
1&5
1
0.22410
146
0.26998
726
0.26919
95
1.27051
503
0.32441
786
1
0.26441
843
0.28025
198
0.29670
482
0.78811
841
0.28034
952
0.290621
95
0.26
8396
94
1
0.34615
068
0.36045
055
0.40768
655
0.79593
673
0.36399
808
0.385938
61
0.36
8518
55
SVM: Radial
Model
Parameters:
0.1
0.34615068
1
0.28736
992
0.29048
576
0.43504
493
0.78647
162
0.30105
254
0.321011
65
0.29
4961
22
SVM: Radial
Model
Parameters:
0.1
0.28736992
SVM:
Polynomial
Model
Parameters:
3&3
0.30570146
SVM: Radial
Model
Parameters:
0.1
0.30562686
10
0.30570
146
0.31876
652
0.37349
433
1.21413
613
0.32010
724
0.337232
66
0.32
3916
46
3
0.30562
686
0.31137
251
0.40285
867
0.75861
262
0.32191
936
0.345044
4
0.31
6775
01
19
Backhoe Loader
- 16.0 + Ft
Standard
Digging Depth
Polyn
omial
Model
Param
eters:
6&1
Hydraulic
Excavator, Track
- 28.0 to 33.0
Metric Tons
Radial
Model
Param
eters:
0.01
Track Type
Tractor, Dozer 105.0 to 130.0
Horsepower
Radial
Model
Param
eters:
0.1
Track Type
Tractor, Dozer 160.0 to 190.0
Horsepower
Polyn
omial
Model
Param
eters:
2&1
Backhoe Loader
- 0.0 to 14.0 Ft
Standard
Digging Depth
Radial
Model
Param
eters:
0.1
Skid Steer
Loader - 0.0 to
701.0 Lb
Operating
Capacity
9
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
9
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Radial
Model
Param
eters:
0.1
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Track Type
Tractor, Dozer 190.0 to 260.0
Horsepower
Radial
Model
Param
eters:
0.01
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Wheel Loader 135.0 to 150.0
Horsepower
Radial
Model
Param
eters:
0.1
7
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Wheel Loader 175.0 to 200.0
Horsepower
Radial
Model
Param
eters:
0.1
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
Wheel Loader 225.0 to 250.0
Horsepower
Polyn
omial
Model
Param
eters:
6&2
9
0.0004
5109
Linear
Regres
sion
Skid Steer
Loader - 2701.0+
Lb Operating
Capacity
Radial
Model
Param
eters:
0.01
9
0.0004
5109
Wheel Loader 90.0 to 100.0
Horsepower
Radial
Model
Param
eters:
0.1
8
Track Type
Tractor, Dozer 260.0 +
Horsepower
Radial
Model
Param
eters:
0.1
10
1&4
1&5
1&5
SVM:
Polynomial
Model
Parameters:
6&1
0.28914711
2
0.28914
711
0.29907
637
0.30620
912
1.24188
101
0.30235
597
0.307575
19
0.30
3117
62
1
0.37914
877
0.39569
984
0.47466
733
0.70126
707
0.39296
672
0.405767
37
0.38
7588
14
SVM: Radial
Model
Parameters:
0.01
0.37914877
1
0.31184
375
0.31313
502
0.41287
421
1.81875
93
0.33867
579
0.339487
76
0.31
9167
43
SVM: Radial
Model
Parameters:
0.1
0.31184375
SVM:
Polynomial
Model
Parameters:
2&1
0.38919826
1
0.38919
826
0.42608
39
0.44200
895
0.94395
754
0.41650
617
0.432248
04
0.41
3769
04
4
0.32020
462
0.33946
19
0.37386
66
0.74734
629
0.35639
792
0.344605
73
0.32
5790
11
SVM: Radial
Model
Parameters:
0.1
0.32020462
0.1&5
2
0.39193
912
0.42895
705
0.48740
001
1.57397
779
0.44306
527
0.421871
85
0.41
3769
01
SVM: Radial
Model
Parameters:
0.1
0.39193912
1&5
2
0.45868
538
0.48185
645
0.52125
878
0.88971
728
0.48800
686
0.486872
52
0.41
3769
Neural
Networks:
1&5
0.1&4
1
0.33270
004
0.33326
997
0.42667
388
0.86321
417
0.34307
985
0.381239
53
0.35
1424
51
SVM: Radial
Model
Parameters:
0.1
0.33270004
GA
M
1&5
1
0.38363
821
0.37672
165
0.45806
171
0.61917
316
0.37705
636
0.402778
14
0.41
3769
KNN: 10
0.37672165
Ridge
Regres
sion
GA
M
1&4
2
0.37865
485
0.34669
888
0.39429
239
1.03299
897
0.37540
148
0.368214
07
0.38
4812
25
KNN: 9
0.34669888
Linear
Regres
sion
Ridge
Regres
sion
GA
M
1&2
1
0.20840
118
0.20688
864
0.22058
635
1.56057
389
0.21495
708
0.222718
23
0.21
7478
48
KNN: 9
0.20688864
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
0.1&1
1
0.30473
316
0.28435
215
0.37389
757
0.90813
827
0.29837
22
0.353921
67
0.32
7818
21
KNN: 8
0.28435215
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
1
0.38991
36
0.40325
283
0.47156
16
0.73189
076
0.40337
091
0.438172
09
0.39
9978
38
SVM: Radial
Model
Parameters:
0.1
0.1&4
0.1&4
1&5
20
0.413769
0.3899136
Motorgrader 200.0 +
Horsepower
Polyn
omial
Model
Param
eters:
2&1
Hydraulic
Excavator, Track
- 5.0 to 6.0
Metric Tons
Radial
Model
Param
eters:
0.1
Wheel Loader 250.0 to 275.0
Horsepower
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
9
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Radial
Model
Param
eters:
0.1
10
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Backhoe Loader
- Unidentified
Radial
Model
Param
eters:
0.01
7
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
Wheel Loader 100.0 to 110.0
Horsepower
Polyn
omial
Model
Param
eters:
5&1
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
Hydraulic
Excavator, Track
- 11.0 to 12.0
Metric Tons
Polyn
omial
Model
Param
eters:
4&2
6
10
8
0.0004
5109
Linear
Regres
sion
Ridge
Regres
sion
GA
M
GA
M
GA
M
1&5
4
0.44490
182
0.45012
806
0.51265
445
0.84326
766
0.47455
712
0.508644
77
0.41
3769
Neural
Networks:
1&5
SVM: Radial
Model
Parameters:
0.1
0.27534269
0.413769
2
0.27534
269
0.28041
916
0.37614
394
1.14462
962
0.30552
916
0.304181
77
0.29
4165
72
0.01&3
1
0.37001
197
0.39039
829
0.41221
317
0.95634
534
0.38010
83
0.408472
65
0.39
9544
75
SVM: Radial
Model
Parameters:
0.1
0.37001197
0.01&3
1
0.26369
973
0.26213
523
0.31400
174
1.27351
349
0.26561
423
0.290554
32
0.30
2389
73
KNN: 7
0.26213523
0.362194
26
0.34
7168
44
SVM:
Polynomial
Model
Parameters:
5&1
0.33089216
0.31
7747
7
SVM:
Polynomial
Model
Parameters:
4&2
0.28398376
0.1&2
1&3
0.1&3
2
1
0.33089
216
0.28398
376
21
0.34598
567
0.34997
783
0.36821
579
0.36778
163
1.29839
931
0.75537
583
0.34628
508
0.35620
213
0.332211
35