Forecasting Students' Success to Graduate Using Predictive Analytics

Abstract

Predictive analytics is the process of forecasting outcomes based on historical data. Execution of predictive analytics involves data collection, analysis and massaging, identifying machine learning, predictive modeling, predictions, and monitoring. Among phases, data analysis and massaging or data preprocessing plays a vital role in the prediction’s result. This study aims to predict the student’s probability of graduating on time using the student’s demographic profile, previous academic achievement, and college admission result. The dataset was acquired from Caraga State University with 2207 samples of new entrants. This study implemented KNN to impute numerical data, while mode imputation was used for categorical values. Moreover, binary encoding was employed for nominal data to prevent algorithm ranked the values in order. Seven (7) algorithms were tested to the original dataset and compared to datasets integrated with LASSO (L1), Ridge (L2) regression, and Genetic Algorithm (GA) separately. The result shows that L1 with Decision Tree classifier has the lowest accuracy (58%) and AUC score (50%). It also has the smallest number of features selected (5). GA, on the other hand, selected thirty-three (33) features with AUC score of 71% and predicted 79% accurately using the Logistic Regression classifier. It exhibited 21% increase in AUC score compared to no feature selected dataset (NFS) with the same classifier.