Al-Hikmah University Journal

Accurate prediction of Master's program eligibility from Computer Science Bachelor's

performance is vital. Despite K-nearest neighbours' (KNN) common use in predictions,

there's a gap in comparing it with the Logistic Regression Classifier (LRC). This study aimed

to address this gap by identifying the most suitable classifier between LRC and KNN for

accurately predicting students' performance in the computer science programme. In order to

evaluate the performance metrics of two classification algorithms, LRC and KNN were

modelled through 10-fold cross-validation in WEKA, with a comprehensive evaluation of

performance metrics for each classifier. The study used secondary data from Al-Hikmah

University, Ilorin, Nigeria (2009-2015) on computer science students' academic

performances. It included 7 attributes and 478 instances for each, comprising three

categorical and four numeric features. Class labels Yi (YES, NO) reflected meeting minimum

admission requirements, with grade scales for class labels including 1.0-1.49(pass), 1.50-

2.3(Third class honor) 2.40-3.49 (Second class honor lower division), 3.5-4.49 (Second class

honor upper division) and 4.5-5.0 (First class honor). LRC showcased superior performance

over KNN, when tuning parameter k = 1with Euclidean distance used as distance metrics,

across multiple metrics, including accuracy (94.7699% vs. 89.9582%), precision (96.1% vs.

92.7%), recall (96.9% vs. 93.8%), F-measure (96.5% vs. 93.3%), ROC Area (97.5% vs.

85.6%), and error rate (5.2301% vs. 10.0418%). Notably, KNN exhibited faster processing

time (0.01 sec vs. 0.07 sec) when compared to LRC. The optimal KNN configuration for the

model was observed when k = 3. The study recommends utilizing LRC as the preferred

predictive model for students' performance in a computer science programme.