SHANU OLANIYI RILWAN

Computer Science

At the Computer Science department office

Appointment on Visitation important

Topic: Expert System Using Fuzzy Logic FCM

Description: My research interest and focus is on Intuitionistic fuzzy similarity measure which gives Boolean value 0-1 but considers both membership and non-membership functions. This has been used to resolve classification problems but with shortcomings. The interest in this area is to research into resolving some of these lapses. Also, my interest on Neural Network, Data Science and Analysis; including application of statistical, mathematical modeling and machine learning analyses for solving real life complex problems.

#	Certificate	School	Year
1.	M.Sc (Computer Science)	Computer Science, Lagos State University, Ojo Lagos.	2018

PREDICTING AND EVALUATING STUDENTS’ ACADEMIC PERFORMANCE USING SMOTE ENHANCED CAR-DT MODEL WITH SHAP INTERPRETABILITY

Academic performance evaluation in educational data mining has emerged as a significant aspect given its relevance and importance not only in predicting student academic development but also enhancing learning results and aiding in institutional decision making. Lecturers in higher institutions tend to use attendance, continual assessments, examinations and other admission-related factors to assess students’ academic aptitude. The challenge however lies in being able to identify students who will likely succeed or fail academically. Often times students are ignorant of their potential graduation status while institutions are unable to provide timely assistance for students who are academically struggling. This study therefore addresses the need to apply an enhanced machine learning technique as a paradigm in the modeling of students’ performance into categorization of good standing or stragglers. We adopted the Classification and Regression Tree algorithm (CAR-DT) and the Synthetic Minority Oversampling Technique (SMOTE) to address the class imbalance inherent in the dataset prior to the model training. The SHapley Additive exPlanations (SHAP) values were computed to give post-hoc interpretability of model predictions at global as well as individual student level, overcoming the back-box constraint of machine learning models in educational situations. The results obtained from our model achieved 90% accuracy, 93% specificity and 85% sensitivity with a positive predicted value (PPV) of 88% and a negative predicted value (NPV) of 91%. The model also achieved an AUC-ROC of 0.94 confirming a string discriminatory capability. The findings revealed that age, travel time, relationship status, alcohol consumption, and previous semester grades are principal predictor variables of academic performance. The SHAP analysis further revealed that the previous semester grades (G1, G2) and daily alcohol consumption (Dalc) have the most significant impact on predicted outcomes. The findings in this study will help departmental heads make targeted decisions, offer appropriate counseling and support to struggling students in their early years.

SHANU RILWAN is a Lecturer II at the Department of Computer Science

SHANU has a M.Sc in Computer Science from Computer Science, Lagos State University, Ojo Lagos.