A Baseline Comparative Study of Supervised Machine Learning Algorithms for Japanese Email Spam Detection

This letter presents a baseline comparative study of six supervised machine learning algorithms—Linear Support Vector Machine, SVM with RBF kernel, Decision Tree, Random Forest, Naive Bayes, and Logistic Regression—for detecting spam in Japanese email. Feature vectors were constructed using MeCab‐based morphological segmentation and Bag‐of‐Words representations. We systematically varied part‐of‐speech configurations, training data volume, and spam–ham ratios. Results indicate that noun‐dominant features yield superior discrimination and that dataset scale and class balance strongly govern performance.