Designing And Implementation Of An Inferential Engine Prototype For Degree-Program Recommendation

ABSTRACT

In this study students’ admission forms and the Internet were used to create datasets in order to design, implement and test a prototype for an inferential engine for degree programme recommendation. The said engine is a Machine Leaning (ML) tool to be used by students for selection of degree programmes. The dataset had 17 features which represented ordinary and advanced level performances, category of schools of admission and student gender. Data in the dataset was unevenly distributed in nine classes, whereby 80% was used for training with 10-fold cross validation and 20% was used for testing of the seven selected ML algorithms.

The ML algorithms which were selected for this study were Decision Tree (ID3), Nearest Neighbor, Support Vector Machine (RBF kernel), and Bagging Classifier. Others were Random Forest, Adaptive Boost, and Neural Network (MLP). Random Forest outperformed the other ML algorithms with an accuracy of 66%, with Mean Absolute Error of 11.93. RF attained precision, recall and F-measure of 66% each; Cohen’s kappa and MCC of 60% each; Log Loss of 29%; and Hamming Loss of 34%.

The study recommends the educational governing institutions to use a well-formed evaluation and record keeping system to enable easy tracking of student performance.