Suspicious Defaulter Forecasting Using Machine Learning Classifiers

Author Name(s): *Ashmeet Singh, Pranav Sharma, Prof. L. Shalini
Author Email: ashmeet.singh2013@vit.ac.in

Abstract

Every year individuals join certain University College filled with diverse set of people which deems to be a stepping stone towards the road one wants to go on.  The aim of any University/ College is to create an environment for individuals to study and succeed in their life. Achieving this involves various activities. One of the most important involved in this process is to identify and guide the defaulters to the path of success. Identification of suspicious defaulters among thousands of individuals is a very big task itself. Such a problem holds the key to the success or downfall of the University/ College. Survey was carried out to collect large amount of data for analysis. On manual analysing the given data 23 factors were found to be contributing towards the result. Then the data was further analysed through five classification algorithms to check the result through various aspects. Among the various algorithms, the most accurate algorithm, Decision Tree, was selected for implementation purpose. The algorithms provided the result in form of ranking as ok, doubtful and very doubtful. Every student must be categorized into one of these ranks and certain set of actions associated to them could be carried by the management to maintain discipline in campus.

Decision Tree algorithm is further coded in python. The python coding act as a backend for the front end of the created web application. Frontend web application provides with an individual suspicion and list of suspicious defaulter which is found by analysis and validation of individual data. The security is maintained through a login page. Such solution ease out the process of suspicion for the College/ University and makes the process faster. It also helps them to guide a niche crowd and not as a whole. This helps the University/ College both qualitatively and quantitatively.

Keywords

Decision Tree, RapidMiner, Naive Bayes, Random Forests, K-NN, Neural Networks.

Introduction

For ensuring a good reputation of the university, a quick identification of defaulters is required. Currently the identification of the defaulters is operated in a non-digital or semi-digital manner. With the rise of machine learning techniques, a technical solution has to be proposed to make the manual process faster, niche, effective, productive and automated.

The proposed system is a web application that views database of defaulters. This list is generated by a prediction algorithm that acts best among all algorithms for the desired dataset. The parameters, such as accuracy, kappa and F-measure, ensures an unbiased and a clear judgement over the classifiers. The dataset was created using the survey of students and close inspection of experts. The data represents the dependencies of various aspects of student details for being involved in forbidden activities. The main focus of discussion will be the selection of best classifier with appropriate tests and analysis conducted on rapidminer tool. The best classifier will be further be used in implementation of the web application. The developed application can play a vital control in easing down the management process. Students will be compelled to maintain an appropriate rank along with the academic grades. Section II gives the overview of previous related works.  The detailed explanation of Dataset and its attributes is given in Section III. Short discussion and the approaches of all the classifiers are given in Section IV. In next section, experiments and analysis are presented and then followed by results and conclusion of the discussed approach.

Conclusion

The system supports the adaption of new system which makes the management process easier. A new competitive approach could be build which will enhance and motivate students to be morally and socially respectable.

Feature selection could further be improved and different strategies could be involved for more influencing attribute generation (e.g. Medical checkup every month for better accuracy of prediction).

The system gives the rudimentary steps towards the future grading system for keeping check over the students. In future, the process could be made as the grade awarding process by applying grade values (A+, A, B+, B, D) by further distribution of values (Okay, Doubtful and Very doubtful) of label attribute (Defaulter list).

394 total views, 1 views today

Download PDF File

About the author: tej