DefinePK

Abstract

Technical education is becoming more and more career-oriented. Therefore, most of the researchers and studies have contributed their part by predicting students’ career. On the other hand, students’ career after graduation has become a major factor in building the reputation of the institutes. Predicting their degrees and future direction beforehand can help to take timely action by institutions. Data mining is a process of finding patterns and correlations within large datasets to predict outcomes. When the data mining process is used to study the learning characteristics, behaviour and performance of the student, it is called Educational Data Mining (EDM). This study implements a supervised machine learning technique to predict career of IT applicants whether IT or Non-IT by analyzing the data of those students’ who had graduated. The dataset is collected from the Course Management System (CMS) of Barani Institute of Information Technology (BIIT) PMAS, AAUR Rawalpindi and mapped with the alumni dataset which is collected from BIIT graduated students using Google forms to have one dataset for experimental use. The major concerns of this work is to develop an efficient student recommendation system for predicting career of BSCS/BSIT applicants’ whether IT or Non-IT at the time of admission and enhance the performance of learning model by re-sampling of dataset with SMOTE algorithm. This work also highlights the impact of selection of most relevant set of features for accurate results. The system efficiency has been tested upon the data of 3327 students. The total number of 11 attributes have been considered for career prediction i.e. gender, current degree program, demography details, SSC board, SSC subject, SSC grade, HSSC board, HSSC subject, HSSC grade, final degree CGPA and alumni job. To improve accuracy of learning models, re-sampling of dataset technique is sused to handle class imbalance problem by applying SMOTE algorithm. In this research, Random Forest, C4.5(J48) and Support Vector Machine (SVM) models are used to determine the best predictive model for supervised machine learning. 10-fold cross validation and standard performance evaluation measures such as: accuracy, precision, recall, F-Measure are used to evaluate the classifiers results. During the experiments Random Forest obtained 96.86% accuracy measure which is better than all of the learning algorithms, i.e., C4.5(J48) 96.55% and SVM 96.15% after handling class imbalance problem with SMOTE. This recommendation system may be more helpful by predicting career of IT department applicants’ at the time of enrollment.