Analisis Sentimen Gugurnya Anggota TNI dalam Misi UNIFIL pada Twitter Menggunakan Perbandingan Model Machine Learning

Silvi Ayuni (Universitas Bina Sarana Informatika)
Panny Agustia Rahayuningsih (Universitas Bina Sarana Informatika)
Muhammad Rezki (Universitas Bina Sarana Informatika)

Jurnal Nasional Komputasi dan Teknologi Informasi
Vol. 9 No. 4 (2026): Agustus, 2026

Abstrak - Gugurnya empat prajurit Tentara Nasional Indonesia (TNI) dalam misi perdamaian United Nations Interim Force in Lebanon (UNIFIL) di Lebanon Selatan pada 29-30 Maret 2026 dan 24 April 2026 memicu reaksi opini publik yang luas di platform Twitter. Penelitian ini bertujuan untuk menganalisis sentimen masyarakat Indonesia terhadap peristiwa gugurnya 4 anggota TNI,  sekaligus untuk membandingkan performa antara lima algoritma machine learning, yaitu Logistic Regression, Random Forest, Support Vector Machine (SVM), Neural Network (MLP), dan Decission Tree. Data dikumpulkan menggunakan Tweet Harvest dengan 11 kata kunci pencarian pada periode 30 Maret hingga 19 Mei 2026.11 File tersebut kemudian di blending dan  menghasilkan 1.632 cuitan yang kemudian setelah melalui tahapan preprocessing data yang dihasilkan menjadi 764 data. Pelabelan data dilakukan secara otomatis menggunakan pendekatan lexicon-based dengan kamus InSet (Indonesian Sentiment Lexicon), menghasilkan distribusi 78% data berlabel negatif dan 22% positif. Ekstraksi fitur menggunakan TF-IDF, serta penanganan ketidakseimbangan kelas menggunakan SMOTE yang diterapkan hanya pada data latih. Pembagian data menggunakan rasio 80:20 dengan Stratified K-Fold Cross Validation (K=5). Hasil evaluasi menunjukkan bahwa Neural Network (MLP) merupakan model terbaik dengan F1-score macro tertinggi sebesar 0,7795, akurasi 0,8627, dan AUC-ROC 0,8485. Kata kunci : Analisis Sentimen; Machine Learning; UNIFIL; Twitter; Imbalanced Data; Abstract - The fall of four soldiers of the Indonesian National Armed Forces (TNI) during the United Nations Interim Force in Lebanon (UNIFIL) peacekeeping mission in South Lebanon on March 29–30, 2026, and April 24, 2026, sparked widespread public opinion reactions on the Twitter platform. This study aims to analyze Indonesian public sentiment regarding the incident of the four TNI members' fall, while also comparing the performance of five machine learning algorithms, namely Logistic Regression, Random Forest, Support Vector Machine (SVM), Neural Network (MLP), and Decission Tree. Data were collected using Tweet Harvest with 11 search keywords during the period from March 30 to May 19, 2026. The 11 files were then blended, yielding 1,632 tweets, which after going through the data preprocessing stages resulted in 764 data points. Data labeling was performed automatically using a lexicon-based approach with the InSet (Indonesian Sentiment Lexicon) dictionary, producing a distribution of 78% negatively labeled data and 22% positively labeled data. Feature extraction used TF-IDF, and class imbalance was handled using SMOTE applied only to the training data. Data splitting used an 80:20 ratio with Stratified K-Fold Cross Validation (K=5). The evaluation results show that the Neural Network (MLP) is the best model with the highest macro F1-score of 0.7795, accuracy of 0.8627, and AUC-ROC of 0.8485. Keywords: Sentiment Analysis; Machine Learning; UNIFIL; Twitter; Imbalanced Data;