International Journal of Engineering Trends and Technology (IJETT) – Volume 29 Number 2 - November 2015
Modeling for Prediction of Tomato Yield and Its Deviation
using Artificial Neural Network
Syed Abou Iltaf Hussain#1, DigantaHatibaruah*2
1
ME student,Department of Mechanical Engineering, Jorhat Engineering College, Jorhat, Assam, India
Associate Professor, Department of Mechanical Engineering, Jorhat Engineering College, Jorhat, Assam,
India
mm2/plant respectively in the leaf area index subI. INTRODUCTION
model. In 2007, B. J I et al have developed a better
Prediction is the way things will occur in the
model than the multiple linear regression-based yield
future. Prediction is a difficult job especially in
models to predict the Fujian rice yield of the Fujian
agriculture predicting the crop yield. As the yield is
province of China from the location-specific rainfall
dependent upon various factors varying from weather
data, soil fertility data and the weather variables such
to the amount of fertilizers required. Out of the
as sunshine hours per day, solar radiation per day and
various factors some factor changes from day to day
temperature sum per day. The values of R2 and
like the temperature of the day, rainfall, sunshine etc.
RMSE are comparatively higher and lower
On the other hand some factors remains constant like
respectively in case of ANN rice yield model than
the pH of soil but some factors like the amount of
multiple linear regression-based yield models. In
fertilizers used depend upon the will of farmer how
2010, Rahman and Balamodeled a network to predict
much they are applying.
the jute production from i) Julian day, ii) solar
According to FAOSTAT, India is the second
radiation, iii) maximum temperature, iv) minimum
largest
producer
of
tomatoes
producing
temperature, v) rainfall and vi) type of biomass. Jiří
approximately 12 million tonnes in 2010 but the
ŠŤASTNÝ et al in 2011 has predicted the crop yield
value rose to 17.5 million metric tonnes of tomatoes
level using artificial neural network. The input values
annually in 2014. Out of all the states growing
were density of nurslings per meter square and
tomatoes, Andhra Pradesh leads the tomato growth in
average onion yield. The model was less complex
India covering approximately 35% of the total
than the other existing models and higher accuracy so
production followed by Karnataka. Assam is one of
that the model is easy to use and the prediction is
the lowest tomato growing states of the country
more accurate. In 2014, SaisuneeJabjone and
producing approximately 402.49 thousand tonnes in
SuraWannasang developed a model that could predict
2013.
the rice production of Phimai district, Thailand from
Basically Tomato cultivators of Phesual,
the technique used for irrigation, rice breed, season,
Assam are not economically strong so they cannot
rice-field area and characteristics, cultivation
invest large amount of money in their farming.
technique and damage area.
Moreover the farmers of Assam have small land
The output from the model was compared stepwise
holdings. Their main motive is to increase their
with the linear regression models and the result
production by putting minimum efforts, involving
obtained from the neural network was better than the
less labours and investing minimum money. Hence a
linear regression method.
need for a predicting tool arises that could predict or
give a rough idea about the amount of the tomato
III. OBJECTIVE
yield at the end of the season so that a decision could
The main objective of this paper is to develop a
be made weather to cultivate or not. Artificial Neural
model to predict tomato yield and its deviation from
Network is one of the many computing models that
the maximum possible amount of production from
could predict the live in situation accurately.
the amount of fertilizers used by the farmers, pH of
soil and land available for tomato cultivation for each
II. LITERATURE SURVEY
farmer.
In 2005 Kaul et al used ANN to develop model to
predict the yield of corn and soybeans by considering
IV. METHODOLOGY
various environmental factors as inputs. In 2006
Data collected were based on the interviewed
Ushadaand Murase used artificial neural network to
survey of random tomato cultivators from the Phesual
develop a model that showed relationship of
region of Jorhat district.
minimum temperature Tmin, maximum temperature
Tmax, optimum temperature T opt and ambient
A. Artificial Neural Network
temperature Tamb with the as heat unit accumulation,
Artificial Neural Networks are a family of
relative rate of growth, leaf area index, height of
statistical learning models inspired by the central
moss, mass of moss and temperature stress factor.
nervous system of animals particularly brain. The
The specific leaf area and the ground area had the
brain learns from the past experience. Brain is a
2
best experimental values of 1.498 m /kg and of 28
complex network of neurons which process signals as
2
ISSN: 2231-5381
http://www.ijettjournal.org
Page 102
International Journal of Engineering Trends and Technology (IJETT) – Volume 29 Number 2 - November 2015
received by the sensory organs and asked to react
according to the situation. Similarly artificial neural
networks are a system of interconnected neurons
which exchange messages with each other. There are
some numeric weights at the connections of the
neurons. These numeric weights can be readjusted
through training. Training is the process of learning
new jobs by repeatedly doing a particular job. When
an artificial neural network model is trained the
predicted output obtained is compared with the actual
output and the numeric weights are updated. When a
particular artificial neural model is trained repeatedly
the numeric weights are updated until the predicted
output and actual output are similar or the error
between the two is the least.
The fundamental unit of neural network is
known as neuron. Artificial neural network are
represented by a set of interconnected neurons which
exchanges messages between each other. The
artificial neuron receives one or more inputs and
sums them to produce an output. The inputs of each
node are summed and then weighted. The sum is
passed through an activation function or transfer
function which is a non-linear function.
Mathematically, for neuron k:
uk =
wkj.xj) and
yk=
(uk+bk)
wherexj are input signals and wkjare synaptic
weights of neuron k, ukis the linear combiner output
due to input signals. bk is the bias,
is the
activation function and output signal of the neuronis
theyk. The use of bias bk has the effect of applying an
affine transformation to to the output uk of the linear
combiner in the model of Figure as shown by the
following equation vk= uk + bk
With respect to the weights in the network the
backpropagation methods calculates the gradient of
loss function. The gradient is fed to the optimization
method which in turn uses it to update the weights, in
an attempt to minimize the loss function.
Feedforward backpropagation algorithm
A feedforward neural network consists of
three layers. The first layer is the input layer, second
or the middle layerconsists of one or more hidden
layers and the third is the output layer. Each of the
input layer, hidden layer and the output layer consists
of a number of neurons. Every neuron of the input
layer is connected to the neurons of the first hidden
layer and every neuron of first hidden layer is
connected to the neurons of the second hidden layer
and so on. The neurons of the last hidden layer are
connected to the neurons of the output layer
1. Input layer
3. Output layer
2. Hidden layer
4. Neurons
Figure 2: Typical feedforward neural network
Cascade-forward backpropagation algorithm
A cascadeforward neural network consists of
three layers. The first layer is the input layer, second
or the middle layerconsists of one or more hidden
layers and the third is the output layer. Each of the
input layer, hidden layer and the output layer consists
of a set of neurons. Every neurons of the input layer
is connected to each neuron of the hidden layer and
the output layer. Every neuron of first hidden layer is
connected to the neurons of the second hidden layer
and so on. The neurons of the last hidden layer are
connected to neurons of the output layer.
1.
2.
3.
Inputs
4. Output paths
Sums
5. wij = weights
Transfer function
Figure 1: A basic artificial neuron
Backpropagation algorithm
Backpropagation is a form of supervised
learning. When using a supervised learning method,
the network is provided with both sample inputs and
predicted outputs. The predicted outputs are
compared against the actual outputs for given input.
ISSN: 2231-5381
http://www.ijettjournal.org
Page 103
International Journal of Engineering Trends and Technology (IJETT) – Volume 29 Number 2 - November 2015
1. Input layer
2. Hidden layer
3. Output layer
Figure 3:- Typical cascade-forward neural network
B. Preprocessing of data
It is one of the many steps in data mining. Data
pre-processing is done before the actual processing of
data. It prepares the raw data for further processing.
Data Normalization
Data pre-processing is also known as Data
Normalization. The reason for using feature scaling
method is that the gradient descent converges faster.
Mathematically,the normalized datax/is given by:
Where (la)iis land available for tomato cultivation
with each farmer
tyiis the tomato yield from that particular
land
4.
V. ANALYSIS AND DISCUSSIONS
Data were collected through an interview
survey amongst the professional tomato cultivators of
the Phesual region of Jorhat district. The collected
data were normalized using feature scaling method so
that the network converges faster.
The ANN network consists of 3 layers. The
first layer is the input layer. It consists of five
neurons viz. i) pH of soil, ii) amount of superphosphate used, iii) amount of potash used, iv)
amount of urea used and v) land available for tomato
cultivation. The second layer is the hidden layer. It
consists of single hidden layer. The hidden layer
consists of 10 neurons. The final layer is the output
layer and it consists of two output neurons viz. i)
tomato yield and ii) deviation of tomato yield from
the maximum. Table-I and Table-II represents the
total predicted tomato yield and its deviation obtained
from the feed-forward neural network and cascadeforward neural network respectively.
x/=
wherex is the input data of a parameter.
min(x) is the minimum input data of the parameter.
max(x) is the maximum input data of the parameter.
Deviation of tomato yield
The tomato cultivators are cultivating the
variety Avinash-2. The main characteristic of this
variety is high yield under controlled conditions.
Yield of this variety is about 1200 quintal per
hectare.
Deviation of tomato yield = 1200*(la)i – tyi
TABLE-I
PREDICTED TOMATO YIELD
Farmer
Actual
yield
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
45
200
190
26.25
90
227.5
70
90
85
105
112.5
165
27.5
37.5
40
78.75
100
82.5
35
25
90
325
162.5
157.5
127.5
ISSN: 2231-5381
Predicted yield
CascadeFeed-forward
forward
46.48
51.88
179.77
185.83
178.36
169.90
31.77
33.13
95.22
88.66
228.27
219.79
66.93
65.86
105.03
111.72
82.25
86.80
107.53
114.82
113.21
115.78
164.65
154.61
31.42
25.83
32.20
34.31
34.67
41.98
76.57
75.52
97.33
109.18
68.68
76.09
33.92
27.36
36.04
44.80
91.83
87.26
193.64
317.50
185.28
190.76
157.67
160.63
106.28
113.25
Farmer
Actual
yield
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
52.5
126
44
44
150
75
10
26.25
25
65
26.25
175
225
165
50
61.25
37.5
80
30
255
45
68.75
105
78.75
225
http://www.ijettjournal.org
Predicted yield
CascadeFeed-forward
forward
52.39
51.46
107.53
114.82
34.62
33.87
34.64
32.84
162.00
169.36
67.05
74.76
33.95
20.93
31.72
32.82
34.57
30.79
62.83
62.17
32.73
21.45
162.00
169.36
223.58
238.42
192.79
181.82
41.81
42.65
68.85
66.96
60.22
68.58
86.22
84.64
32.47
28.33
254.74
234.19
34.38
37.89
48.51
60.95
101.07
95.16
74.91
74.01
194.95
183.65
Page 104
International Journal of Engineering Trends and Technology (IJETT) – Volume 29 Number 2 - November 2015
A scatter plot diagram is plotted for tomato yield
by taking actual values along the abscissa and
predicted along the ordinate. The equation of best fit is
given by:
A. For feed-forward neural network:
y = 0.8476x + 11.76
Where y is the predicted tomato yield
x is the actual tomato yield
From the scatter plot diagram following points are
observed:
1. The predicted output showed a linear
relation with the actual output.
2. The co-efficient of determination for the
line of best fit is 0.9685. Hence we can
conclude thatthe predicted output from
the model is having 96.85% accuracy.
PREDICTED
PREDICTED Vs ACTUAL
400.00
350.00
300.00
250.00
200.00
150.00
100.00
50.00
0.00
R² = 0.911
total
yield(quintal)
Linear (total
yield(quintal))
0
200
400
600
ACTUAL
Figure 4:- Line of best fit between the actual and
predicted tomato yield from the feed-forward
backpropagation algorithm
From the scatter plot diagram following
points are observed:
1. The predicted output showed a linear relation
with the actual output
2. The co-efficient of determination for the line
of best fit is 0.911 i.e. the predicted output
from the model is having 91.1% accuracy.
B. For cascade-forward neural network:
y = 0.9476x + 4.6771
where y is predicted tomato yield
x is actual tomato yield
PREDICTED
PREDICTED vs ACTUAL
R² = 0.9685
450.00
400.00
350.00
300.00
250.00
200.00
150.00
100.00
50.00
0.00
total yield(quintal)
Linear (total
yield(quintal))
0
200
400
600
ACTUAL YIELD
Figure 5: Line of best fit between the actual and
predicted tomato yield from the cascade-forward
backpropagation algorithm
ISSN: 2231-5381
http://www.ijettjournal.org
Page 105
International Journal of Engineering Trends and Technology (IJETT) – Volume 29 Number 2 - November 2015
TABLE-II
PREDICTED DEVIATION TOMATO YIELD
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
276.07
602.68
452.15
214.55
552.15
896.26
251.07
391.61
236.07
376.61
690.18
798.22
133.04
203.3
281.07
483.13
702.68
399.11
286.07
Farmer
Actual
deviation of
tomato yield
39
40
41
42
43
44
798.22
351.34
500.63
444.11
562.15
210.8
Predicted deviation of tomato
yield
Feed-forward
Actual
deviation of
tomato yield
Farmer
Cascadeforward
250.05
250.17
627.10
633.43
469.68
486.93
208.91
217.43
540.57
556.38
809.36
934.07
256.46
232.12
357.94
354.88
256.09
217.07
354.27
352.73
681.47
716.58
803.01
848.42
176.70
176.10
207.86
215.61
251.09
263.77
472.50
486.17
694.92
727.51
385.64
387.15
292.85
281.57
Predicted deviation of tomato
yield
CascadeFeed-forward
forward
771.93
802.76
336.78
342.29
476.86
495.56
385.63
395.20
544.08
564.37
219.57
224.67
A scatter plot diagram is plotted for
deviation of tomato yield by taking actual values
along the abscissa and predicted along the ordinate.
The equation of best fit is given by:
a. For feed-forward neural network:y = 0.8083x + 67.813
Where y is the predicted deviation of tomato
yield
x is the actual deviation of tomato yield.
Predicted deviation of tomato
yield
Feed-forward
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
296.07
552.15
1280.36
640.18
1287.33
354.11
429.11
355.61
277.07
277.07
652.68
406.61
150.54
214.55
296.07
256.07
214.55
627.68
979.02
Actual
deviation of
tomato yield
Farmer
45
46
47
48
49
50
708.22
276.07
332.59
456.88
483.13
738.22
Cascadeforward
248.66
260.44
541.84
559.31
816.87
1215.23
621.55
629.82
1095.80
1163.51
356.13
353.81
399.84
409.48
354.27
352.73
268.14
272.69
271.33
273.91
643.59
646.28
385.61
388.62
186.30
185.73
209.30
217.90
278.44
276.51
255.56
236.02
252.64
237.17
643.59
646.28
815.56
994.58
Predicted deviation of tomato
yield
CascadeFeed-forward
forward
726.94
778.18
257.92
268.22
316.59
326.22
472.60
463.78
473.36
488.02
770.58
801.27
PREDICTED vs ACTUAL
PREDICTED DEVIATION
Farmer
Actual
deviation of
tomato yield
1600.00
1400.00
1200.00
1000.00
800.00
600.00
400.00
200.00
0.00
R² = 0.985
Deviation from
optimal (quintal)
0
500
1000
ACTUAL DEVIATION
1500
Linear
(Deviation from
optimal
(quintal))
Figure 6: Line of best fit between the actual and
predicted deviation of tomato yield from the feedforward backpropagation algorithm
From the scatter plot diagram following
points are observed:
ISSN: 2231-5381
http://www.ijettjournal.org
Page 106
International Journal of Engineering Trends and Technology (IJETT) – Volume 29 Number 2 - November 2015
1
b.
Theactual output is showing linear
relation with thepredicted output.
2 As the co-efficient of determination for
the line of best fit is 0.9341. Hence we
can conclude that the predicted output
from the model is having 93.41%
accuracy.
For cascade-forward neural network:y = 0.9835x + 7.2749
Where y is the predicted deviation of tomato
yield
x is the actual deviation of tomato yield
PREDICTED DEVIATION
PREDICTED vs ACTUAL
1400.00
1200.00
1000.00
800.00
600.00
400.00
200.00
0.00
R² = 0.934
Deviation from
optimal
(quintal)
0
500
1000
ACTUAL DEVIATION
1500
Linear
(Deviation
from optimal
(quintal))
predicted deviation showed 96.85%
accuracy.
In case of cascade-forward neural network
following points are observed:
1. The predicted tomato yield and its deviation
are varying linearly with the actual tomato
yield and its deviation respectively.
2. From the scatter plot of tomato yield it was
found that the predicted value of tomato
yield showed 93.41% accuracy and the
predicted deviation showed 98.53%
accuracy.
Cascade-forward neural network can predict the
tomato yield and its deviation more accurately than
the feed-forward neural network as in case of
cascade-forward neural network the input layer is
directly linked with the output layer. As a result of
which the synaptic weights could be better updated
this is not possible in case of feed-forward neural
network. Hence it can be concluded that Cascadeforward neural network of Artificial Neural Network
can be used efficiently and effectively for prediction
of tomato yield and its deviation.
REFERENCES
Temeyer Bradley R. and Gallus William A. Jr. “ Using an
Artificial Neural Network to Predict Parameters for Frost
Deposition on Iowa Bridgeways”
2. Kaul M., Hill R.L., Walthall C., “Artificial neural network
for corn and soybean prediction”, Agricultural System 85
(2005) 1-18.
3. Ushada M., Murase H. “Identification of a Moss Growth
System using an Artificial Neural Network Model”,
Biosystemss Engineering 94 (2) (2006) 179-189.
4. Ji B. , Sun Y., Yang S. and Wan J. (2007), “Artificial neural
networks for rice yield prediction in mountainous regions”,
249-261
5. Rahman M.M., Bala B.K., “Modelling of jute production
using artificial neural networks”, Biosystems Engineering
105 (2010) 350-356.
6. Qiao D.M., Shi H.B., Pang H.B., X.B. Qi, Plauborg F.,
“Estimating plant root water uptake using a neural network
approach”, Agricultural Water Management 98 (2) (2010)
251-260.
7. WANG Xin-Zheng, DUAN Xiao-chen and LIU Jing-yan,
“Application of Neural Network in the Cost Estimation of
Highway Engineering” Journal Of Computers, Vol. 5, No.
11, NOVEMBER 2010, 1762-1766
8. ŠŤASTNÝ Jiří, KONEČNÝ Vladimír, TRENZ Oldřich,
“Agricultural data prediction by means of neural network”,
Agric. Econ. – Czech, 57, 2011 (7): 356–361
9. H. Khan Zabir, S. AlinTasnim and Hussain Md. Akter, “Price
Prediction of Share Market using Artificial Neural Network
(ANN)”, International Journal of Computer Applications
(0975 – 8887) Volume 22– No.2, May 2011, 42-47.
10. Miss.SnehalDahikar
S.,
V.RodeDr.Sandeep
(2014),
“Agricultural Crop Yield Prediction Using Artificial Neural
Network Approach”, INTERNATIONAL JOURNAL OF
INNOVATIVE
RESEARCH
IN
ELECTRICAL,
ELECTRONICS, INSTRUMENTATION AND CONTROL
ENGINEERING Vol. 2, Issue 1, January 2014, 683-686.
1.
Figure 7: Line of best fit between the actual and
predicted deviation of tomato yield from the cascadeforward backpropagation algorithm
From the scatter plot diagram following points are
observed:
1. Theactual output is showing linear relation with
thepredicted output.
2. As the co-efficient of determination for the line of
best fit is 0.9853 i.e. the predicted output from the
model is having 98.53% accuracy
VI. CONCLUSION
The purpose of the study was to model a
network that would predict the tomato yield and its
deviation in a specific region. Two different networks
were created where one was feed-forward and the
other was cascade-forward neural networks. Both the
neural networks were trained and learned by
backpropagation algorithm.
In case of feed-forward neural network
following points are observed:
1. The predicted tomato yield and its deviation
are varying linearly with the actual tomato
yield and its deviation respectively.
2. From the scatter plot of tomato yield it was
found that the predicted value of tomato
yield showed 91.1% accuracy and the
ISSN: 2231-5381
http://www.ijettjournal.org
Page 107
International Journal of Engineering Trends and Technology (IJETT) – Volume 29 Number 2 - November 2015
11. JabjoneSaisunee and WannasangSura (2014), “Decision
Support System Using Artificial Neural Network to Predict
Rice Production in Phimai District, Thailand”, International
Journal of Computer and Electrical Engineering, Vol. 6, No.
2, April 2014, 162-166.
12. TengelengSiddi and Armand Nzeukou (2014), “Performance
of Using Cascade Forward Back Propagation Neural
ISSN: 2231-5381
Networks for Estimating Rain Parameters with Rain Drop
Size Distribution”, www.mdpi.com/journal/atmosphere,
ISSN 2073-4433, 454-472.
13. http://www.cropnutrition.com/efu-soil-ph
14. http://www.indiaagronet.com/tomato/resources/3/seed1.htm
http://www.ijettjournal.org
Page 108