Detecting the Impact of Social Media on Users' Mental Health Using Machine Learning and XAI

Ara Bela Zulfa Laila (Department of Computer Science, Faculty of Mathematics and Natural Sciences, Universitas Negeri Semarang)

Jurnal Buana Informatika
Vol. 17 No. 1 (2026): Jurnal Buana Informatika, Volume 17, Nomor 1, April 2026

This research develops a machine learning-based predictive system to detect potential depression due to social media use, and compares the performance of algorithms such as Random Forest, XGBoost, and Naïve Bayes. Survey data, including age, gender, relationship status, daily usage duration, and social media platform, were used to build the model, with accuracy, precision, recall, and F1-score evaluated. XGBoost showed the best performance with 90% accuracy and a high F1-score. The main features that affect depression prediction include duration of social media use, age, and platforms. Explainable AI (XAI) techniques with LIME increase the transparency of the model, provide relevant explanations for individuals, and strengthen confidence in the predictions. This research emphasizes the importance of transparency in model implementation in the mental health field and offers a flexible solution that can be adopted for digital applications such as chatbots or real-time mental health monitoring dashboards.   Penelitian ini mengembangkan sistem prediktif berbasis machine learning untuk mendeteksi potensi depresi akibat penggunaan media sosial, serta membandingkan kinerja algoritma seperti Random Forest, XGBoost, dan Naïve Bayes. Data survei yang meliputi usia, jenis kelamin, status hubungan, durasi penggunaan harian, dan platform media sosial digunakan untuk membangun model dengan evaluasi akurasi, precision, recall, dan F1-score. XGBoost menunjukkan kinerja terbaik dengan akurasi 90% dan F1-score tinggi. Fitur utama yang memengaruhi prediksi depresi meliputi durasi penggunaan media sosial, usia, dan platform. Teknik Explainable AI (XAI) dengan LIME meningkatkan transparansi model, memberikan penjelasan yang relevan untuk individu, dan memperkuat kepercayaan terhadap prediksi. Penelitian ini menekankan pentingnya transparansi dalam penerapan model di bidang kesehatan mental dan menawarkan solusi fleksibel yang dapat diadopsi untuk aplikasi digital seperti chatbot atau dashboard pemantauan kesehatan mental real-time.