Performance Analysis of KNN and C.45 Algorithms with Kappa Measure Evaluation

EasyChair Preprint 13603

Abstract

This study aims to develop an effective predictive model for forecasting students' graduation status at the Health Polytechnic of the Ministry of Health in Medan. Using the K-Nearest Neighbors (K-NN) method and the C4.5 algorithm, along with applying the SMOTE technique, addresses class imbalance in the dataset and enhances the accuracy of predicting students' graduation. Data were obtained from the Health Polytechnic of the Ministry of Health in Medan, including lists of students who graduated and those who did not from 2018 to 2021. The dataset comprises 3997 records with 15 features, including student ID numbers, activity statuses, Grade Point Averages (GPAs), credit unit counts, and graduation statuses. Class imbalance handling was performed using SMOTE. The KNN and C4.5 models were evaluated with three K values (1, 3, and 5) and three optimal node splitting methods. Model evaluation using the Kappa Measure indicated a high level of agreement between predicted and actual outcomes. Data normalization enhanced the consistency of analysis results, while SMOTE improved the quality of analysis and classification outcomes. Both the KNN and C4.5 models effectively predicted graduation statuses, with KNN showing improved performance with increasing K values. Cohen's Kappa values confirmed high agreement across all models. The study's conclusion underscores the effectiveness of data normalization, SMOTE, and predictive models (KNN and C4.5) in forecasting students' graduation statuses, offering valuable insights to support decision-making in higher education institutions.

Keyphrases: C.45, K-Nearest Neighbors(K-NN), Kappa Measure, Prediksi kelulusan mahasiswa, SMOTE

@booklet{EasyChair:13603,
  author    = {Dina Mariana Siboro and Suyanto Suyanto and Amalia Amalia},
  title     = {Performance Analysis of KNN and C.45 Algorithms with Kappa Measure Evaluation},
  howpublished = {EasyChair Preprint 13603},
  year      = {EasyChair, 2024}}