Presentation - University of Wisconsin–Madison

advertisement
ECE 539 Course Project
NEURAL NETWORK APPROACHES
FOR
AUTOMOBILE MPG PREDICTION
12/14/2010
Xiaofei Sun
University of Wisconsin-Madison
Motivations

Nowadays, fuel economy becomes a great concern of the
governments and drivers

MPG varies with vehicle specs and conditions


Database available online only accounts for different models

Large amount of data required
Build NN models to predict the MPG based on given specs
and conditions

MLP

RBF
1/8
Data Description

Source: UCI Machine Learning Repository
http://archive.ics.uci.edu/ml/datasets/Auto+MPG

8 Inputs:
1. cylinder #
2. displacement
3. horsepower
4. weight
5. acceleration
6. year
7. origin
8. manufacturer

1 Output: MPG
2/8
Data Preparation

392 sets of data

Correlation coefficients between I/O were calculated
0.9
Correlation Coefficient
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
3/8
Linear Regression
7-way cross validation

Training MSE = 11.12
Tuning MSE = 12.70
50
45
45
40
35
y = 1.1164x - 1.8062
R² = 0.8476
35
30
30
Target
Target
40
y = 1.0058x - 0.142
R² = 0.8335
25
20
25
20
15
15
10
10
5
5
0
0
0
10
20
Output
30
40
0
10
20
Output
30
40
4/8
Multi Layer Perceptron

MATLAB Neural Network Toolbox Used

Learning algorithms:


Gradient descent with momentum

Scaled conjugate gradient

Levenberg-Marquardt 
Datasets were randomly divided into three subsets:

60% for training

20% for validation (early stopping)

20% for testing
5/8
Multi Layer Perceptron
Structure: 7-12-1 feedforward network

Log-sigmoid function for hidden layer

Linear function for output layer
Test MSE = 5.11
Training MSE = 4.03
50
45
40
35
30
25
20
15
10
5
0
y = 1.0055x - 0.1154
R² = 0.9346
Target
Target

0
10
20
30
Output
40
50
45
40 y = 0.9775x + 0.5947
35
R² = 0.9157
30
25
20
15
10
5
0
0
10
20
30
Output
40
50
6/8
Conclusions and Future Work

MLP yields better performance than linear regression after
fine tuning
14
12
10
8
6
MLP
4
Linear Regression
2
0
Training MSE
Test MSE

Will construct radial basis function network, and compare
with MLP
7/8
Any Questions?
8/8
Download