Recently, intelligent soft computational techniques such as
Artificial Neural Network (ANN), Fuzzy Inference System (FIS) and (ANFIS) can
model superiority of human knowledge features. They also re-establish the
process without plenty of analysis. Thus these techniques are attracting great
attention in an environment that is obvious with the absence of a simple and
well-defined mathematical model. Besides, these models are characterized by
nonrandom uncertainties which associated with imprecision and elusiveness in
real-time systems. Many researchers have studied the application of neural
networks to overcome most of the problems above outlined.
The fuzzy set theory
is also used to solve uncertainty problems.
The use of neural nets in
applications is very
sparse due to its implicit
the prohibitive computational effort and so on. The key benefit of fuzzy
logic is that its
knowledge representation is explicit, using
simple IF-THEM relations. However, it is
at the same time its major limitation. The Attrition Rate Prediction
cannot be easily described
unknown parameters. The integration of neural network into the fuzzy logic system makes it possible to learn from
the prior obtained data sets.
The purposes of this study are to compare the applicability of ANN and ANFIS in
predicting Attrition Rate in an Organization and to identify the most fitted model
to the study area.
The input data used for Attrition Rate prediction are the different employee characteristics and this
data is acquired by Kaggle, an open
source dataset platform.
This graph presents the correlations between each variables. The size of
the bubbles reveals the significance of the correlation, while the color
present the direction (either positive or negative).
Artificial neural network (ANN)
A customized neural network is adopted here. A network first needs
to be trained before interpreting
information. Several different algorithms are available
for training of neural networks, but the back-propagation algorithm is the most versatile
learning procedure for multilayer neural networks. Also, the fact that back-propagation algorithms are
especially capable to solve problems of prediction
makes them highly popular.
of the network, data are processed through the network until they
reach the output layer. In
output is compared to
the measured values. The difference or error between the two is processed back through the network (backward pass) updating the individual weights
of the connections and the biases of the individual
neurons. The input and output data are mostly represented as vectors called training pairs. The process as mentioned
above is repeated for all the training
pairs in the data set, until the network error has
converged to a threshold minimum
defined by a corresponding cost function, usually the root mean squared
neural network is used for predicting Attrition Rate. A number of 15,000 data e.g. were
utilized during training session and 50 data
e.g. were used during testing session. A suitable configuration has to
be chosen for the best performance of the
network. Out of the different configurations
tested, two hidden layer with 50 and 25 hidden neurons
produced the best result. The log sigmoid function was employed as an activation function.
Suitable numbers of epochs have to be assigned to overcome the problem of over fitting
and under fitting of data.
ANN structure for
Neuro Fuzzy Inference System (ANFIS)
ANFIS was originally proposed by JSR Jang. ANFIS is a fuzzy system trained on the set of input and output data by an
algorithm derived from the theory of Artificial Neural Networks. The algorithm is a hybrid training
algorithm based on back
propagation and the least squares approach. In
this algorithm, the parameters defining the shape of the membership functions are identified
by a back
while the consequent parameters
are identified by the least squares method. An ANFIS can be viewed as a three- layer feed forward neural network.
The first layer one represents input
variables, the layer
two represents fuzzy rules, and
the layer 3 is an output.
For ANFIS model, similar training and testing data sets were
used as in ANN model. We used Subtractive
Clustering algorithm in ANFIS for training the dataset.
Comparison of ANN and ANFIS models
Results from two models are presented in this section to access and compare the
degree of prediction accuracy and generalization capabilities of the two networks designed in the present problem. The same training and testing data sets were used to
train and test both models to extract more solid conclusions from the comparison results.
Mean square error (MAE), root
mean square error (RMSE)
were calculated based on the corresponding measured data. Analysis of data in randomized sets clearly
showed that ANN model is best fit for predicting the Attrition Rate.
In this paper we
showed the ability of ANFIS and the ANN in predicting the Attrition Rate and
potential candidates who are going to leave the firm.
The results showed that the
RMSE, MSE for the training data were 0.088, 0.007 for the ANN model, and 0.160,
0.025 for the ANFIS model. As for unseen data, the RMSE, MSE were 0.8, 0.89 for
the ANN and 0.5, 0.25 for the ANFIS model. The ANFIS model, however, was more
sensitive than the ANN model for the unseen data set and is performing better
for the same.
We can conclude
that ANN model can fit the output better compared to the ANFIS model for the
unseen data set. But ANFIS is better than ANN in generalization and prediction
of unseen data.