Implementasi SMOTE pada Klasifikasi Email Spam Menggunakan Algoritma Machine Learning Berbasis TF-IDF

Moh Ali Aljauhari (Universitas Negeri Yogyakarta)
Fatchul Arifin (Universitas Negeri Yogyakarta)

Jurnal Algoritma
Vol 23 No 1 (2026): Jurnal Algoritma

Class imbalance is a major challenge in email spam detection, causing classification models to be biased toward the majority class. This study examines the effectiveness of the Synthetic Minority Over-sampling Technique (SMOTE) on six machine learning algorithms—Naive Bayes, SVM, KNN, Logistic Regression, Random Forest, and XGBoost—using TF-IDF feature representation. Test results show that synthetic data balancing successfully improved the sensitivity of linear models (SVM and Logistic Regression) with a recall value exceeding 49.1%, overcoming the failure to predict the minority class in the original dataset. KNN recorded the highest F1-score (0.436), while Random Forest provided the best class separation stability with an AUC-ROC of 0.601. The main contribution of this study is to demonstrate that although SMOTE improves minority class detection capabilities, its effectiveness on high-dimensional text data remains limited by feature sparsity constraints that trigger class overlap.