Wabah penyakit campak di Indonesia, termasuk Kabupaten Sumenep, menimbulkan beragam opini masyarakat melalui media sosial X (Twitter). Analisis sentimen digunakan untuk mengidentifikasi kecenderungan sentimen pada tweet terkait wabah campak. Penelitian ini bertujuan menganalisis sentimen dan juga membandingkan kinerja ekstraksi fitur Term Frequency–Inverse Document Frequency (TF-IDF) dan Bag of Words (BoW) menggunakan algoritma Logistic Regression, serta mengetahui pengaruh penerapan Synthetic Minority Over-sampling Technique (SMOTE) terhadap kinerja klasifikasi. Data diperoleh melalui crawling pada platform X menggunakan kata kunci terkait wabah campak, dengan 6.144 tweet yang dikumpulkan pada periode Oktober 2024 hingga Oktober 2025. Rangkaian penelitian meliputi pengumpulan data, preprocessing teks, pelabelan sentimen, pembagian data, ekstraksi fitur TF-IDF dan BoW, klasifikasi Logistic Regression, serta evaluasi menggunakan accuracy, precision, recall, dan F1-score. Hasil pengujian tanpa SMOTE menghasilkan TF-IDF memperoleh accuracy 76,8%, weighted precision 76,9%, weighted recall 76,8%, dan weighted F1-score 76,8%, sedangkan BoW memperoleh accuracy 74,8%, weighted precision 75,1%, weighted recall 74,8%, dan weighted F1-score 74,9%. TF-IDF unggul 2,0 poin persentase berdasarkan accuracy. Setelah SMOTE, TF-IDF dan BoW memperoleh accuracy yang sama, yaitu 76,6%. Penerapan SMOTE meningkatkan accuracy BoW sebesar 1,8 poin persentase, sedangkan TF-IDF mengalami penurunan 0,2 poin persentase. Dengan demikian, TF-IDF dengan Logistic Regression tanpa SMOTE menghasilkan kinerja tertinggi dengan accuracy 76,8%.


Deskripsi Lain

The measles outbreak in Indonesia, including Sumenep Regency, has sparked diverse public opinions on the social media platform X (Twitter). Sentiment analysis was employed to identify sentiment trends in tweets related to the measles outbreak. This study aimed to analyze sentiment, compare the performance of Term Frequency–Inverse Document Frequency (TF-IDF) and Bag of Words (BoW) feature extraction methods using the Logistic Regression algorithm, and determine the impact of applying the Synthetic Minority Over-sampling Technique (SMOTE) on classification performance. Data were obtained by crawling the X platform using keywords related to the measles outbreak, resulting in a dataset of 6,144 tweets collected between October 2024 and October 2025. The research process encompassed data collection, text preprocessing, sentiment labeling, data splitting, TF-IDF and BoW feature extraction, Logistic Regression classification, and performance evaluation using accuracy, precision, recall, and F1-score metrics. Testing results without SMOTE showed that TF-IDF achieved an accuracy of 76.8%, weighted precision of 76.9%, weighted recall of 76.8%, and weighted F1-score of 76.8%, whereas BoW achieved an accuracy of 74.8%, weighted precision of 75.1%, weighted recall of 74.8%, and weighted F1-score of 74.9%. TF-IDF outperformed BoW by 2.0 percentage points in terms of accuracy. Following the application of SMOTE, both TF-IDF and BoW achieved an identical accuracy of 76.6%. The application of SMOTE increased BoW's accuracy by 1.8 percentage points, while TF-IDF experienced a decrease of 0.2 percentage points. Consequently, the combination of TF-IDF and Logistic Regression without SMOTE yielded the highest performance, achieving an accuracy of 76.8%.