Perbandingan SVM dan Logistic Regression untuk Klasifikasi Teks Konsultasi Daring Kehamilan dan Menstruasi

Abstract

The high volume of doctors' response texts in online health consultations, particularly on pregnancy and menstruation topics, has increased the need for an accurate and automated medical text classification system. However, these two topics often share similar medical terminology and consultation contexts, making the classification process challenging. This study aims to develop an automatic classification model for Indonesian-language doctors' responses in online health consultations by comparing two algorithms, namely Support Vector Machine (SVM) and Logistic Regression (LR), which were selected because they are widely used in text classification tasks with TF-IDF feature representation and are capable of handling high-dimensional data. The contributions of this study include the development of a classification model using an underexplored Indonesian doctors' response dataset, feature importance analysis to identify dominant medical terms in each category, and a comparative evaluation of SVM and LR performance validated using McNemar's statistical test. The study utilized the public "Doctor's Answer Text Dataset in Indonesian" from Mendeley Data, comprising 17,807 records. The models were developed using the CRISP-DM framework with TF-IDF feature extraction and evaluated across four train-test split ratios using accuracy, precision, recall, and F1-score metrics. The results showed that Logistic Regression (LR) consistently outperformed Support Vector Machine (SVM), achieving the highest accuracy of 0.9359 with an 80:20 train-test split. Feature importance analysis identified "hamil" (pregnant) and "janin" (fetus) as the dominant features for the pregnancy category, and "menstruasi" (menstruation) and "siklus" (cycle) for the menstruation category. These findings demonstrate that TF-IDF-based LR is an effective approach for classifying Indonesian-language online health consultation texts.