brain_3_1
63
Time-Delay Artificial Neural Network Computing Models for Predicting Shelf
Life of Processed Cheese
Sumit Goyal
Senior Research Fellow, National Dairy Research Institute, Karnal – 132001- India
thesumitgoyal@gmail.com
Gyanendra Kumar Goyal
Emeritus Scientist, National Dairy Research Institute, Karnal – 132001- India
gkg5878@yahoo.com
Abstract
This paper presents the capability of Time–delay artificial neural network models for
predicting shelf life of processed cheese. Datasets were divided into two subsets (30 for training and
6 for validation). Models with single and multi layers were developed and compared with each
other. Mean Square Error, Root Mean Square Error, Coefficient of Determination and Nash -
Sutcliffo Coefficient were used as performance evaluators, Time- delay model predicted the shelf
life of processed cheese as 28.25 days, which is very close to experimental shelf life of 30 days.
Keywords: Time – delay, ANN, Artificial Intelligence, Processed Cheese, Shelf Life,
Prediction
1. Introduction
This paper highlights the significance of Artificial Neural Network (ANN) models for
predicting shelf life of processed cheese stored at 30
o
C. ANN is inspired by the functional aspects
of biological neural networks. A neural network consists of an interconnected group of artificial
neurons, and it processes information using a connectionist approach to computation. In most cases
an ANN is an adaptive system that changes its structure based on external or internal information
that flows through the network during the learning phase [1].Processed cheese is very nutritious and
generally manufactured from ripened Cheddar cheese, but sometimes less ripened Cheddar cheese
is also added in lesser proportion. Its manufacturing technique includes addition of emulsifier, salt,
water and sometimes selected spices. The mixture is heated in jacketed vessel with continuous
stirring in order to get homogeneous mass. It is a protein rich food. This variety of cheese has
several advantages over raw and ripened Cheddar cheese, such as excellent supplement to meat
protein, tastier with longer shelf life.
Time-Delay Neural Network (TDNN) is an alternative neural network architecture whose primary
purpose is to work on continuous data. The advantage of this architecture is to adapt the network
online and hence helpful in many applications. The architecture has a continuous input that is
delayed and sent as an input to the neural network [2]. In this study, TDNN’s models with single
and multilayer layers for predicting shelf life of processed cheese have been developed.
Single Layer
Single layer perceptron network consists of a single layer of output nodes; the inputs are fed
directly to the outputs via a series of weights. The sum of the products of the weights and the inputs
is calculated in each node, and if the value is above some threshold (typically 0) the neuron fires
and takes the activated value (typically 1); otherwise it takes the deactivated value (typically -1).
Neurons with this kind of activation function are also called artificial neurons or linear threshold
units [3].
Multilayer
This class of networks consists of multiple layers of computational units, usually
interconnected in a feed-forward way. Each neuron in one layer has directed connections to the
neurons of the subsequent layer. In many applications the units of these networks apply a sigmoid
function as an activation function. Multilayer networks use a variety of learning techniques, the
BRAIN. Broad Research in Artificial Intelligence and Neuroscience
Volume 3, Issue 1, February 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print)
64
most popular being back-propagation. Here, the output values are compared with the correct answer
to compute the value of some predefined error-function. By various techniques, the error is then fed
back through the network. Using this information, the algorithm adjusts the weights of each
connection in order to reduce the value of the error function by some small amount. After repeating
this process for a sufficiently large number of training cycles, the network usually converge to some
state where the error of the calculations is small [3].
Shelf Life
Shelf life is the length of time given before a product is considered unsuitable for sale, use,
or consumption. In some regions, a best before, use by or freshness date is required on packaged
perishable foods. Use prior to the expiration date does not necessarily guarantee the safety of a food
or drug, and a product is not always dangerous or ineffective after the expiration date [4]. ANNs
have been applied for predicting shelf life of Kalakand [5], milky white dessert jeweled with
pistachio [6], and instant coffee flavoured sterilized drink [7, 8]. Time-Delay and Linear Layer
ANN models were developed for predicting shelf life of soft mouth melting milk cakes [9], soft
cakes [10]. Radial Basis models were successfully applied for predicting shelf life of Brown milk
cakes [11]. The aim of this study is to develop TDNN single and multilayer models for predicting
shelf life of processed cheese stored at 30
o
C, and to compare the developed models with each other.
The outcome of this research would be very useful for cheese manufacturers, retailers, consumers
and researchers.
2. Method Material
The data consisted of 36 samples, which were divided into two subsets, i.e., 30 used for
training the network and 6 for testing the TDNN models. Soluble nitrogen, pH, standard plate
count, yeast & mould count, and spore count were taken as input parameters, and sensory score as
output parameter for developing TDNN single and multilayer models (Fig.1).
Figure 1. Inputs and output parameters for TDNN models
Soluble nitrogen
pH
Standard plate count
Yeast & mould count
Spore count
Sensory
Score
S. Goyal, G. K. Goyal - Time-Delay Artificial Neural Network Computing Models for Predicting Shelf Life of Processed
Cheese
65
Many combinations were tried and tested, as there is no defined rule of getting good results
rather than hit and trial method. As the number of neurons increased, the training time also
increased. Several algorithms like Bayesian regularization, Levenberg Marquardt algorithm,
Gradient Descent algorithm with adaptive learning rate, Powell Beale restarts conjugate gradient
algorithm and BFG quasi-Newton algorithms were tried. Backpropagation algorithm based on
Bayesian regularization was finally selected for training the networks, as it gave most promising
results. TDNN was trained up to 100 epochs with single as well as multiple hidden layers. Transfer
function for hidden layer was tangent sigmoid while for the output layer, it was pure linear
function.
Figure 2. Training pattern of TDNN models
The Neural Network Toolbox under MATLAB software was used for developing the TDNN
models. Training pattern of TDNN models is presented in Fig.2.
Weights Selection
Error Evaluation
Weights Adjustment
Training TDNN models
Training
Dataset
Error Calculation
Minimum
Error
End
No
Yes
BRAIN. Broad Research in Artificial Intelligence and Neuroscience
Volume 3, Issue 1, February 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print)
66
Measures for Prediction Performance
−
= ∑
2
1
exp
N
cal
n
QQ
MSE
(1)
−
= ∑
2
1 exp
exp1
N
cal
Q
QQ
n
RMSE
(2)
−
−= ∑
2
1
2
exp
exp2
1
N
cal
Q
QQ
R
(3)
−
−
−= ∑
2
1 expexp
exp2
1
N
cal
QQ
QQ
E
(4)
Where,
expQ = Observed value;
cal
Q = Predicted value;
expQ =Mean predicted value;
n = Number of observations in dataset.
Mean Square Error: MSE (1), Root Mean Square Error: RMSE (2), Coefficient of Determination:
R
2
(3) and Nash - Sutcliffo Coefficient:E
2
(4) were used in order to compare the prediction ability of
the developed models.
3. Results and discussion
TDNN model’s performance matrices for predicting sensory scores are presented in Table 1
and Table 2, respectively.
Table 1. Performance of single layer for predicting sensory score
Neurons MSE RMSE R2 E2
3 3.01226E-05 0.005488403 0.994511597 0.999969877
5 0.000128522 0.011336741 0.988663259 0.999871478
6 0.00019086 0.013815219 0.986184781 0.99980914
8 0.000177938 0.013339332 0.986660668 0.999822062
10 1.20278E-06 0.001096714 0.998903286 0.999998797
12 2.41185E-05 0.004911058 0.995088942 0.999975882
S. Goyal, G. K. Goyal - Time-Delay Artificial Neural Network Computing Models for Predicting Shelf Life of Processed
Cheese
67
14 3.46851E-05 0.005889404 0.994110596 0.999965315
15 0.00021437 0.014641378 0.985358622 0.99978563
16 0.001314906 0.036261628 0.963738372 0.998685094
18 0.000840727 0.028995291 0.971004709 0.999159273
20 8.5198E-05 0.009230276 0.990769724 0.999914802
24 8.91481E-06 0.002985769 0.997014231 0.999991085
30 9.62368E-05 0.009810037 0.990189963 0.999903763
Table 2. Performance of multilayer for predicting sensory score
Neurons MSE RMSE R2 E2
3:3 4.67356E-05 0.006836347 0.993163653 0.999953264
4:4 2.77834E-05 0.005270993 0.994729007 0.999972217
5:5 3.40619E-05 0.00583626 0.99416374 0.999965938
7:7 0.000428285 0.020695048 0.979304952 0.999571715
8:8 2.18493E-05 0.004674322 0.995325678 0.999978151
9:9 0.000160167 0.012655697 0.987344303 0.999839833
10:10 1.3025E-05 0.003609012 0.996390988 0.999986975
11:11 0.000266194 0.016315438 0.983684562 0.999733806
12:12 4.47103E-05 0.006686576 0.993313424 0.99995529
13:13 0.000111797 0.010573389 0.989426611 0.999888203
14:14 0.000112257 0.01059513 0.98940487 0.999887743
15:15 0.00022446 0.014981988 0.985018012 0.99977554
16:16 9.56217E-05 0.009778634 0.990221366 0.999904378
TDNN single and multilayer models were developed; for single layer TDNN model with 10
neurons, the best performance was MSE: 1.20278E-06, RMSE: 0.001096714, R2: 0.998903286,
E2: 0.999998797; while for the multilayer combination of 10:10 neurons MSE: 1.3025E-05,
RMSE: 0.003609012, R2: 0.996390988, E2: 0.999986975 performed the best. On comparing them
with each other, it was observed that single layer TDNN model performed better. Therefore, it was
selected for predicting the shelf life of processed cheese. The comparison of Actual Sensory Score
(ASS) and Predicted Sensory Score (PSS) for single layer and multilayer models are illustrated in
Fig.3 and Fig.4, respectively.
BRAIN. Broad Research in Artificial Intelligence and Neuroscience
Volume 3, Issue 1, February 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print)
68
Figure 3. Comparison of ASS and PSS single layer model
Figure 4. Comparison of ASS and PSS for multilayer model
R
2
was found to be 96.5 percent of the total variation as explained by sensory scores. Period
of storage (days) for which the processed cheese has been in the shelf can be determined based on
sensory score (Fig. 5).
S. Goyal, G. K. Goyal - Time-Delay Artificial Neural Network Computing Models for Predicting Shelf Life of Processed
Cheese
69
Figure 5. Sensory score and period of storage for processed cheese
The shelf life is calculated by subtracting the obtained value of days from experimentally
determined shelf life, which was found to be 28.25 days. The predicted value is slightly higher than
the experimentally obtained shelf life of 30 days.
4. Conclusion
TDNN models with single and multi layers were developed taking soluble nitrogen, pH,
standard plate count, yeast & mould count, spore count as input parameters, and sensory score as
output parameter for predicting the shelf life of processed cheese stored at 30
o
C. Mean Square
Error, Root Mean Square Error, Coefficient of Determination and Nash - Sutcliffo Coefficient were
used in order to compare the prediction ability of the developed TDNN models. Regression
equations were developed for predicting the shelf life of processed cheese, which came out as 28.25
days. Since, predicted value is close to the experimentally determined shelf life of 30 days, hence
from the study it can be concluded that TDNN artificial neural network models are quite efficient in
predicting shelf life of processed cheese.
References
[1] http://en.wikipedia.org/wiki/Artificial_neural_network (accessed on 19.8.2011)
[2] http://en.wikipedia.org/wiki/Time_delay_neural_network (accessed on 30.8.2012)
[3] R.A. Chayjan. ”Modeling of sesame seed dehydration energy requirements by a soft-
computing”. Australian Journal of Crop Science, vol. 4, no.3, pp.180-184, 2010
[4] http://www.answers.com/topic/shelf-life (accessed on 30.8.2011)
[5] Sumit Goyal and G.K. Goyal. ”Advanced computing research on cascade single and double
hidden layers for detecting shelf life of Kalakand: An artificial neural network approach”.
International Journal of Computer Science & Emerging Technologies, vol.2, no.5, pp.
292- 295, 2011.
[6] Sumit Goyal and G.K. Goyal. “A new scientific approach of intelligent artificial neural
network engineering for predicting shelf life of milky white dessert jeweled with pistachio”.
International Journal of Scientific and Engineering Research, vol.2, no. 9, pp.1-4, 2011.
[7] Sumit Goyal and G.K. Goyal. ”Cascade and feedforward backpropagation artificial neural
networks models for prediction of sensory quality of instant coffee flavoured sterilized
BRAIN. Broad Research in Artificial Intelligence and Neuroscience
Volume 3, Issue 1, February 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print)
70
drink”. Canadian Journal on Artificial Intelligence, Machine Learning and Pattern
Recognition. vol. 2, no. 6, pp.78-82, 2011.
[8] Sumit Goyal and G.K. Goyal. ”Application of artificial neural engineering and regression
models for forecasting shelf life of instant coffee drink”. International Journal of Computer
Science Issues, vol. 8, no.4, pp. 320- 324, 2011
[9] Sumit Goyal and G.K. Goyal. ”Development of intelligent computing expert system models
for shelf life prediction of soft mouth melting milk cakes”. International Journal of
Computer Applications, vol.25, no.9, pp.41-44, 2011.
[10] Sumit Goyal and G.K. Goyal. ”Simulated neural network intelligent computing models for
predicting shelf life of soft cakes”. Global Journal of Computer Science and Technology.
vol.11, no. 14, version 1.0, pp. 29-33, 2011.
[11] Sumit Goyal and G.K. Goyal. “Radial basis artificial neural network computer
engineering approach for predicting shelf life of brown milk cakes decorated with almonds”.
International Journal of Latest Trends in Computing. vol. 2, no.3, pp. 434-438, 2011.