Integrating Machine Learning Tool

to Improve DSS Design

R. G. Joshi and H. S. Fadewar

Abstract This paper describes how a machine learning tool can be applied to deci-
sion support system. We have used fuzzy logic to enhance performance of DSS. Fur-
ther, system developed is implemented in agriculture domain for selection of suitable
crop. Selection of crop is complex process as it involves number of parameters where
uncertainty is more common for example rainfall, suitable seeds, fertilizers, number
of soil parameters, temperature, air quality, humidity, and so on. The present work
focuses on soil parameters and few other parameters which support proper growth
of crops. Fuzzy logic is applied to those parameters for handling data uncertainty.
This is an attempt to suggest proper decision and reduce the burden by designing
new DSS. Experimental set-up shows increased crop production up to 10–12%.

Keywords Machine learning · Fuzzy logic · Decision support system

1 Introduction

Machine learning is a ability to make computer automatically learn and improve

the quality of solution from previous experiences [1]. There are varieties of situa-
tions handled effectively by the machine learning algorithms like speech recognition,
pattern matching, and so on. There are several machine learning tools like Neural
Network called Artificial Neural Network, Fuzzy Logic, and Genetic Algorithm for
handling different types of programming situations. Principal component analysis
and independent component analysis may are also used as an analysis tool due to
their accuracy [2]. In proposed system, we have used fuzzy logic. Because we want
to process certain linguistic variables and fuzzy logic is best choice for it.

R. G. Joshi · H. S. Fadewar (B)

School of Computational Science, S.R.T.M. University, Nanded, MS, India
e-mail: fadewar_hsf@yahoo.com
R. G. Joshi
e-mail: yma.joshi@gmail.com

© Springer Nature Singapore Pte Ltd. 2019 837

B. Iyer et al. (eds.), Computing, Communication and Signal Processing,
Advances in Intelligent Systems and Computing 810,
https://doi.org/10.1007/978-981-13-1513-8_85
838 R. G. Joshi and H. S. Fadewar

2 Decision Support System

Decision-making process is obvious part of any organizational process. Decision

support system are computer-assisted system which helps the decision maker to
arrive at certain decision. According to Herbart Simon [3], process of decision-
making is divided into three parts: identification of problem domain that requires
decision-making called intelligence phase, finding out solutions and their feasibility
called design phase, and selection of best possible alternative called choice phase.
In conventional programming, the program written once becomes rigid. To maintain
flexibility in the program, here we have decided to use fuzzy logic as a machine learn-
ing tool for decision-making process. In recent development, number of technologies
can be observed in DSS design process. Two important techniques under artificial
intelligence are Rule-Based System (RBS) and Case-Based Reasoning (CBR) [4, 5].
RBS do not express the things in declarative and static way rather it applies set of
“if-then” rules to problem under investigation at specific instances. On the contrary,
in CBR techniques, expertise is recorded in library of different cases happened. Nor-
mally, each case is associated with specific problem and solution for it. The proposed
system is built around RBS.

3 Steps Involved in System Development

Overall steps in system development are shown in the following diagram (Fig. 1).

3.1 Design of FIS (Fuzzy Inference System)

The first step in system development is to frame fuzzy system. Fuzzy system frame-
work involves process of conversion of crisp value into fuzzy value [6]. Further,
we have to apply fuzzy operator to antecedent, proposition from the antecedent to
the consequent, aggregation of the consequents across the rules, and defuzzification.
These are the important points one has to consider while developing fuzzy inference
system. Fuzzy inference is the process of formulating the mapping from a given input
to an output using fuzzy logic [7]. There are two types of fuzzy inference systems,
namely mamdani and sugeno type. Mamdani is more commonly used fuzzy infer-
ence system in which singleton output of membership function is expected, whereas
Sugeno type of system are used when output is either linear or constant. For the
proposed system, mamdani type of FIS is used.
Integrating Machine Learning Tool to Improve DSS Design 839

Fig. 1 Research flow

diagram of fuzzy inference Identification of input parameters
system

Selection of proper parameters

Data Collection

Design of Membership Functions

Rule base Design

Implementation and Testing

output

3.2 Selection of Input Parameter

As shown in Fig. 2, input given to the system is crisp or fuzzy data. If data given to the
system is crisp, it must be converted into fuzzy. This process is called fuzzyfication.
For development of present system, fuzzy logic toolbox from MatLab is used for
the entire process. As we want to implement the proposed system in agriculture
domain, by consulting agriculture expert, we have decided parameters which affect
on production. This process is very complex because a user has to consider different
parameters where uncertainty is more common (Fig. 3). For example, atmospheric
condition, rainfall, proper selection of seeds, suitable fertilizers, soil parameters, and
so on. In the proposed system, we concentrate on soil parameters. It is mandatory
for user to test soil form laboratory, so that system will give accurate result. For
simplicity, it was decided to consider majority of soil parameters and few closely
related with it. In this way, 15 different parameters are decided and system is build
around it. The selected parameters are depicted in Fig. 4.

3.3 Membership Function Design

Figure 3 shows membership function editor from MATLAB. Appropriate design of

membership function for input and output variables is done through this section. This
is the further important step in system development.
840 R. G. Joshi and H. S. Fadewar

Fig. 2 Fuzzy inference system (source www.Researchgate.net)

Fig. 3 Membership function editor

As per system requirement, gauss2mf (Gaussian) membership function for fuzzy-

fication of input and trimf (Triangular) membership function for fuzzyfication of
output is used.
Integrating Machine Learning Tool to Improve DSS Design 841

Fig. 4 Input given to system

3.4 Designing Rule-Based System

Once parameters and their membership functions are decided, the next major step in
system design is to write rule-based system. Rules are defined in following manner.
If (pH is High) and (Salwater is Low) and (Acdwater is High) and (WaterLvl is
High) and (Nitrogen is Vhigh) and (Phosphate is Low) and (Potessium is Low) and
(Depthness is Avg) and (Temp is Avg) and (Humidity is Avg) and (Oxygen is High)
and (Sunlight is Avg) and (Iron is Avg) and (Zinc is Avg) and (Maganese is Low),
then (crop is Wheat). This is another example of crop Green gram.
If (pH is Avg) and (Salwater is Avg) and (Acdwater is High) and (WaterLvl is
High) and (Nitrogen is Vhigh) and (Phosphate is Low) and (Potessium is Low) and
(Depthness is High) and (Temp is Avg) and (Humidity is Low) and (Oxygen is High)
and (Sunlight is Avg) and (Iron is Low) and (Zinc is High) and (Maganese is High),
then (crop is Green gram).
In this way, 22 different rules have been defined for the proper output. Writing
such rule with several combinations is lengthy process (Fig. 5).
Figure 4 illustrates the input parameters given to system and their reference values,
e.g., Nitrogen, Phosphate, Potassium, Iron, Zinc, and so on. System gives output in
fuzzy form and defuzzification is done by using centroid method. Finally, Fig. 6
show output generated by the system.
842 R. G. Joshi and H. S. Fadewar

Fig. 5 Output of the system

comparision of production
25

20
Production

15

10

0
2014 2015 2016 2017

Year
sunflower potato

Fig. 6 Comparison of crop productivity

Integrating Machine Learning Tool to Improve DSS Design 843

Table 1 Comparative study of crop production

Sr. no. Year Traditional Production System rec- Production Quintal/acre
crop quintal/acre ommended
crop
Old values New values
1 2014 Sun flower 12 Potato 18 20
2 2015 Sun flower 10 Potato 17 19
3 2016 Sun flower 12 Potato 20 22
4 2017 Sun flower 11 Potato 21 22

4 Result and Discussion

Earlier researchers Sri Hartati and Imas have used fuzzy approach for evaluating
land suitability and selecting crops. But they have considered 10 soil parameters,
whereas we have considered 15 soil parameters [8]. Kefaya et al. [9] used adaptive
fuzzy neuro approach for crop yield production. But considering only single crop,
we covered 25 different crops. Mawle and Chavan [10] used fuzzy system but their
focus is to predict fertility of soil, whereas we provide exact suitable crop. Kaur [11]
used machine learning approach as it covers irrigation and disease management in
Indian agriculture. Our system recommends suitable crop using fuzzy concept.
Traditional system for selection of crop is based on past experiences. Our exper-
imental set-up is based on scientific soil testing. It has been observed that, system
developed so far recommends other crops than the traditional system, and it shows
increased growth in production.
Table 1 shows difference in production of crop from traditional system to proposed
system. It clearly shows increased production of crop. Graph in Fig. 6 indicate the
same. System developed around soil parameter gives suitable crop. According to
agriculture expert, the parameter selected affects 10–12% on total productivity. We
have tested results from traditional system and fuzzy-based system and results from
developed system are found to be satisfactory and shows 10–12% increased growth.

References

1. Ahmad, I.: Fuzzy Logic For Embedded Systems Applications, pp. 113–118, Elsevier India Pvt.
Ltd. (2006)
2. Patil, M., Iyer, B., Arya, R.: Performance evaluation of PCA and ICA algorithm for facial
expression recognition application. In: Proceedings of Fifth International Conference on Soft
Computing for Problem Solving, pp. 965–976 (2016)
3. Jawdekar, W.: Management Information System, pp. 23–47. TMH (2002)
4. Hayes-Roth, F.: Rule-based systems. Commun. ACM 28(9), 921–932 (1985)
5. Ignizio, J.: Introduction to Expert Systems: The Development and Implementation of Rule-
Based Expert Systems. McGraw-Hill, New York (1990)
844 R. G. Joshi and H. S. Fadewar

6. Chakraborty, D., Pal, N.R.: A neuro-fuzzy scheme for simultaneous feature selection and fuzzy
rule-based classification. IEEE Trans. Neural Netw. 15 (2004)
7. Shaofeng, et al.: Integration of Decision Support Systems to Improve Decision Support Per-
formance. Springer, London (2009)
8. Hartati, S., Sitanggang, I.S.: Fuzzy based DSS for evaluating land suitability selecting crop. J.
Comput. Sci. 6(4), 417–424 (2010). ISSN 1549-3636
9. Qaddoum, K., Hines, E., Illiescu, D.: Adoptive neuro-fuzzy modeling for crop yield predic-
tion. Recent Research in Artificial Intelligence, Knowledge Engineering and Database, ISBN
978–960-474-273-8 (2014)
10. Mawle, Chavan, V.: Fuzzy inference system for productivity and fertility of soil. Int. J. Eng.
Dev. Res. 2(3), 2321–9939 (2014)
11. Kaur, K.: Machine learning: application in indian agriculture. Int. J. Adv. Res. Comput. Com-
mun. Eng. 5(4), 2278–1021 (2016)