Role and Performance of Different Traditional Classification and Nature-Inspired Computing Techniques in Major Research Areas

In the last few years, different machine learning techniques such as supervised, unsupervised, and reinforcement learning have been effectively employed to solve distinct real-life multidisciplinary problems. These techniques have been effectively applied to accurately predict the problems related to stock values, disease diagnosis, sentiment analysis, text processing, gene classification, crop prediction, and weather forecasting. The objective of this manuscript is to present the systematic review on the use of these techniques in five major domains i.e. agriculture, finance, healthcare, education and engineering. A standard review methodology has been adapted to include and exclude the related literature. The performance of different supervised and nature-inspired computing techniques have been accessed on the basis of different performance metrics. The publication trend on the use of machine learning techniques in these five research areas has been also explored. Finally, the gaps in the study have been identified that will assist prospective researchers who want to pursue their research in these areas.


Introduction
Machine Learning(ML) is one of the major multidisciplinary research areas. As per Stanford University, Machine Learning is defined as a science that makes computer to perform some intelligent activities based upon existing data and without being explicitly programmed [1]. Machine Learning has been used in a wide sphere of life. There are three main categories of machine learning called supervised learning, unsupervised learning and reinforcement learning [2]. These techniques have been effectively used to solve a wide variety of classification, clustering and prediction problems. In supervised learning, the inputs are labeled and these labels are the desired outputs. These techniques assist in the data classification process. Disease diagnosis, stock prediction, sentiment analysis are some of the major application areas for supervised learning techniques. Traditionally, different techniques like naïve Bayes, decision tree, random forest and support vector machine have been dominantly used to solve different data classification problems. On the contrary, an unsupervised learning technique also known as clustering techniques deals with unlabeled data. No extra information is provided for grouping the data that's why these are called unsupervised techniques. Reinforcement learning is concerned with the behavior of software agents. There are some functions associated with these agents and they perform their operations in the specified environment to achieve the reward [3].
In the last few years, exponential growth in the use of machine learning techniques has been observed. Different machine learning techniques have been Samriti Sharma, G. Singh, D. Singh 2 employed to solve variety of problems related to engineering [4], finance [5], medical science [6] [7], weather forecasting [8], education [9], transportation [10], robotics [11] and agriculture [12]. Figure1 represents major applications of these techniques.

Figure1: Applications of Machine Learning
The intention of this research work is to present a systematic review of the role and performance of different machine learning techniques in five major research areas viz. agriculture, finance, healthcare, education, and engineering. Moreover, the publication trend of supervised and nature-inspired computing techniques used in these five research areas have been explored to determine the rate of publication of these two techniques in solving the real-life problems of these five major domains. After exploring different indexing databases, it has been found that several review articles have been written on the use of machine learning in different applications. Some of the major objectives of this study are: -To briefly introduce the machine learning techniques.
-To accentuate the least and the most explored research domains by machine learning techniques. -To present and analyze the role of machine learning in five major research areas(agriculture, finance, healthcare, education, and engineering) -To explore the general publication trend of machine learning techniques. -To find and present the rate of publication of natureinspired computing techniques in agriculture, finance, healthcare, education, and engineering. Section 2 represents related works. Review methodology is presented in Section 3. The role and performance of machine learning techniques in five different areas are given in Section 4. Section 5 depicts the publication trend of related articles. Finally, the concluding remarks are presented in Section 6.

Related Works
In the last few decades, several review articles on the machine learning techniques and their applications have been published. Some of the studies are briefly introduced below: Kim et al. [13] have reviewed ten different machine learning based manuscripts. However, they haven"t specified any inclusion/exclusion criteria for their study. No significant findings and suitable future directions have been mentioned. Liakos et al. [12] have studied more than a hundred manuscripts related to the use of ML techniques in agriculture. It is an extensive study including the brief introduction of various classifiers and an explicitly stated review methodology. Subhadra Mishra et al. [14] have presented a review paper on the applications of machine learning techniques in crop production. No significant findings and future directions for the prospective researchers have been highlighted. Fan Cai et al. [15] have surveyed distinct clustering techniques for the analysis of financial data. Lin et al. [5]have reviewed different related articles to determine the data mining techniques which are periodically used in distinct business applications. Future directions are explicitly mentioned. Vivek Rajput et al. [16] have reviewed only eighteen articles related to stock market prediction using data mining and sentiment analysis. No significant details are generated in the study. Wei-Yang Lin et al [17] have performed an extensive review on the various machine learning techniques for bankruptcy prediction and credit scoring. Significant findings are provided by the authors for the practitioners which are interested to pursue their research in the related field. Ashish Sharma et al. [18] have discussed only regression techniques for stock market prediction. No significant future suggestions are mentioned. Kaur and Sharma [19] have presented an extensive analysis of data mining and soft computing techniques for mining diabetic patients. A systematic approach was used while selecting and filtering the article for their review. The authors concluded that in the last decade, a significant rise in the use of data mining and soft computing techniques for early diagnosis of diabetes has been found. Additionally, there is still a need for smart and intelligent diabetes diagnostic framework. Kaur and Sharma suggested combining machine learning and soft computing techniques with the Internet of Things, ontology and information theory for more precise diabetic classification results. Shubham Bind [20] [21]have proposed big data and machine learning based diagnostic system model. There are four layers of the model. ML layer is responsible for disease diagnosis. Data security layers assist in providing security to the data. Different security techniques like activity monitoring, granular access control, PAM, OTP etc. have been used. Data storage layer provides the facility to store different types of data. Data source layer provide different source for data analysis. Divya Tomar et al. [7] have explored the various applications of data mining approaches in healthcare domain. Authors have explained different classification and clustering techniques along with their merits and de-merits and brief summary of machine learning techniques used in healthcare applications and the future directions for the other researchers are also presented. Shweta H.Jambukia et al. [22] have presented a detailed review on the classification of ECG(Electrocardiogram) signals. Authors have also discussed the various ECG databases, pre-processing techniques and issues involved in ECG classification. They have observed that neural networks give better results for ECG classification. Meherwar Fatima et al. [2] have presented a contigent analysis of different machine learning techniques for dignosing five different diseases viz. Diabetes, heart disease, dengue, hepatitis and liver disease. They also highlighted the merits and de-merits of machine learning techniques used in diagnosisng thes e diseases. Sharma et al. [6] have critically examined the role and performance of different data mining techniques used in different lifestyle based human disorders. Authors extensively surveyed more than eighty manuscripts. Authors concluded that a lot of mining work has been carried out for diabetes and cardio problems. However, as a little attention has been paid to develop a predictive model for the diseases viz. ophthalmology, dentistry, and digestive disorders. Therefore, there is a need to explore or mine these areas.

Review Methodology
In this study, a systematic review methodology has been adopted. Different articles related to machine learning and their use in agriculture, finance, healthcare, education, and engineering has been explored by executing different queries on Google Scholar. The study covers more than a hundered articles. The articles embodied in this study have been extracted from peculiar indexing journals such as, IEEE, Elsevier, Springer, Plos|one etc The results found by executing the queries have been scrutinized and filtered based upon title and abstract. Additionally, the final decision regarding inclusion/exclusion has been made based upon the complete contents of the manuscripts. While exploring articles some of the restrictions were applied: -The language was restricted to English only. -Only conference and journal articles were considered.
-Patents and other secondary sources have been ignored

Review Results
This section will highlight the role of supervised and nature-inspired computing(NIC) techniques in agriculture, finance, healthcare, education, engineering etc.

Supervised Learning and Nature Inspired Computing Techniques
Supervised learning techniques are important machine learning techniques that assist in data classification. Inspite of classification supervised learning techniques are also employed in programming or prediction problems. Stock forecasting, crop prediction, disease diagnosis, gene classification are some of the important applications of these techniques. Number of techniques such as Naive Bayes, Decision Tree, SVM, CART(Classification and Regression Tree) and regression have been designed and deployed for the same [1].
Nature Inspired Computing (NIC)Techniques are stochastic techniques which have been inspired from individual or swarm behaviour of various human beings, animals, birds and other natural phenonmenon like wind, water as well as universe.There are several NIC techniques. Some of most admired are GA(Genetic algorithm), ABC(Artificial Bee Colony),ALO(Ant Lion Optimization), FFA(Firefly Algorithm), GWO

Role of Machine Learning in Agriculture
Machine learning has become a vital approach in the field of agriculture satisfying a number of objectives. These techniques have drastically changed the traditional way of performing agricultural activities. In last few years, it has been observed that a variety of supervised and NIC techniques have been used in predicting the wheat yield, weed detection, boosting crop yield, soil management, crop disease detection etc. ML techniques are also playing a vital role in yield prediction which is considered as the most important objective of agricultural planning. ML approach has been also used in modeling the river suspended sediment which is an important concern in managing the water resources [34]. Moreover, the ML techniques are extensively used in analyzing the agricultural data diagnosing crop diseases. In recent years, ML approaches have been employed for estimating

Role of Machine Learning in Finance
Finance is another important research area for machine learning experts and data scientists. Stock and risk analysis are two major subareas of finance. In the business world, the stock market greatly affects the economic advancement of a country. A country"s economic growth is directly proportional to the performance of the stock market. Stock markets generate ample amount of data so different machine learning methods are used for finding the hidden patterns from the data and predicting the stock prices, future trends which helps the investors to make investments in stock market. In spite of stock and risk analysis, loan approval, bankruptcy are other important research areas. Business enterprise is full of risks viz.market risk, credit risk and operation risk etc. so banks have to implement different policies to handle these risks. Machine learning techniques seem to be very useful for risk management [41]. Several  (KSE) and concluded that Ada-Boost, MLP and Bayesian Network give better outcomes than others. Ashish Sharma et al. [18] have surveyed different regression techniques viz. Polynomial regression, RBF(Radial Basis Function) regression, Sigmoid regression and Linear regression and concluded that Linear Regression is considered as a better one in making the predictions than others. Regression analysis is basically used in finding the cause and effect relationships between dependent and independent variables. Yuqinq He et al. [51] have calculated the twelve technical indicators for analyzing the stock market trend. Authors have also studied the three different feature selection algorithms viz. Principle Component Analysis (PCA), Genetic Algorithm (GA) and Sequential Forward Selection(SFS) along with their merits and de-merits.

Role of Machine Learning in Education
Machine Learning(ML) techniques hold a great obligation for education. These techniques have been employed in the prediction of student dropout in higher education. [52]. ML techniques are also applicable in performing the student modeling. Dursun Delen [9] has used SVM decision trees, neural networks, and logistic regression and ensemble methods and observed that ensemble methods perform better than others for predicting the student's retention management. S. Kotsiantis et al. [53] have employed ensemble classifiers for the prediction of student "s performance in distance education which allows the teachers to understand the fact that which students will complete their course of study and which will not. Getting placed in a renowned company is a dream of every student and they work hard to achieve their goal. Every reputed organization has a placement cell for selecting the potential students and improving their skills. ML techniques have been also employed in this domain. Sentkil Kumar Thangave et al. [54] have presented a recommendation system for the student"s placement analysis an achieved an accuracy of 71.66%. Authors have used five different data mining techniques viz. Naïve Bayes, MLP, Reptree, J48, decision tree etc. for predicting the academic performance of students. A.Pavithra et al. [55] have proposed a framework based on five different machine learning techniques viz. Decision tree, Naive Bayes, J48, predicting the academic performance of students. Authors have also mentioned the various socio-economic, non-educational factors that affect the academic conduct of students viz. semester percentage annual income of student, higher secondary marks, the occupation of father, locality of students, psychology etc. Pena-Ayala et al. [56] have presented an extensive survey of various data mining techniques used for the student modeling. Authors have also discussed computer-based education systems as an substitute to the traditional education systems. Various flaws and strengths of educational data mining are also mentioned.

Role of Machine Learning in Engineering
In the last few years, it has been noticed that machine learning is providing viable alternatives to the traditional methods of solving the engineering problems in different domains such as software engineering, civil engineering, chemical engineering, mechanical engineering, computer engineering etc. Jagath Sri Lal Senanayaka et al.

Role of Machine Learning in Healthcare
Nowadays, a large number of people are suffering from different human disorders such as diabetes, cancer, cardio, neuro digestive, and psychological disorders. A huge amount of data related to medical diagnosis is available, so it is required to classify the whole data to make predictions about the diseases and        Table 3 depicts the publication trend of different nature-inspired algorithms in five research areas viz. agriculture, finance, healthcare, education, engineering. It has been observed that ample amount of work has been published in these domains using Genetic Algorithms(GA), Ant Colony Optimization(ACO) and Artificial Bee Colony(ABC) algorithm and a relatively lesser number of articles have been published using other algorithms. Additionally, little attention has been paid to use of multiverse optimization in five major domains(agriculture, finance, healthcare, education, engineering).

Conclusion
This paper presents the systematic review of role and performance of machine learning techniques in five major research areas viz. healthcare, education, finance, agriculture, and engineering. Some of the major applications areas of ML techniques are highlighted. The research works of some of the key authors related to the use of ML techniques particularly in agriculture, finance, education, engineering, and healthcare has been examined and presented in this study. Furthermore, to examine the rate of publication, the publication trend of the related articles has been analyzed. From the last ten year of publication trend, it is observed that a significant amount of research work has been carried out for exploring the role and performance of different ML techniques in engineering. However, agriculture, finance, and healthcare still need more attention. Additionally, as far as nature inspired computing (NIC) techniques are concerned, more attention is required for multiverse optimization techniques. Moreover, the latest and emerging NIC techniques should also be employed in these areas and their performance need to be examined.