Claim Missing Document
Check
Articles

Found 33 Documents
Search

Klasifikasi Multi-Class Pada Kondisi Kesehatan Mental Berbasis Data Media Sosial Menggunakan Model BERT erlin elisa; Sunardi; Herman Yuliansyah
Jurnal Komputasi Vol. 14 No. 2 (2026): Jurnal Komputasi
Publisher : Jurusan Ilmu Komputer Fakultas MIPA Universitas Lampung

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.23960/komputasi.v14i2.368

Abstract

Kesehatan mental merupakan kondisi kesejahteraan psikologis yang memungkinkan individu mengelola stres, bekerja secara produktif, dan berinteraksi secara sosial. Peningkatan penggunaan media sosial menghasilkan data teks yang dapat merepresentasikan kondisi emosional dan psikologis pengguna. Berbagai penelitian telah mengembangkan klasifikasi kondisi kesehatan mental berbasis data media sosial menggunakan pendekatan machine learning dan deep learning. Namun, sebagian besar penelitian masih berfokus pada klasifikasi biner atau deteksi satu kondisi mental tertentu, sedangkan pendekatan multi-class untuk membedakan beberapa kondisi kesehatan mental secara bersamaan masih terbatas. Selain itu, kemiripan pola linguistik antar kondisi mental serta penggunaan bahasa informal pada media sosial menjadi tantangan dalam meningkatkan performa klasifikasi teks tidak terstruktur. Oleh karena itu, penelitian ini mengusulkan model Bidirectional Encoder Representations from Transformers (BERT) untuk mengklasifikasikan kondisi kesehatan mental berbasis data media sosial. Model BERT dipilih karena mampu memahami konteks bahasa secara bidirectional melalui mekanisme self-attention sehingga lebih efektif menangkap representasi semantik dibandingkan metode konvensional. Dataset yang digunakan adalah Reddit Mental Health dengan lima kategori, yaitu Stress, Depression, Bipolar Disorder, Personality Disorder, dan Anxiety. Tahapan penelitian meliputi preprocessing teks, tokenisasi menggunakan BERT tokenizer, pembagian data secara stratified, penanganan imbalanced data menggunakan class weighting dan focal loss, serta fine-tuning model. Hasil eksperimen pada 5.957 data menunjukkan bahwa konfigurasi hyperparameter terbaik memperoleh accuracy sebesar 80%, dengan rata-rata precision, recall, dan F1-score masing-masing sebesar 80%. Kategori Stress memperoleh F1-score tertinggi sebesar 87%, sedangkan kategori Depression sebesar 73%. Hasil penelitian menunjukkan bahwa model BERT efektif dalam memahami pola linguistik dan konteks emosional pada data media sosial tidak terstruktur.
Perbandingan Model Machine Learning untuk Klasifikasi Autism Spectrum Disorder pada Toddler Screening Dataset Nurul Azwanti; Sunardi Sunardi; Herman Yuliansyah
Jurnal Komputasi Vol. 14 No. 2 (2026): Jurnal Komputasi
Publisher : Jurusan Ilmu Komputer Fakultas MIPA Universitas Lampung

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.23960/komputasi.v14i2.370

Abstract

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder characterized by impairments in social communication, social interaction, and repetitive behaviors. Early detection of ASD in toddlers is important to support faster and more appropriate intervention. Previous studies have applied machine learning algorithms, such as Support Vector Machine (SVM), Random Forest, Naive Bayes, and K-Nearest Neighbor (KNN), for ASD classification with relatively high accuracy. However, variations in datasets, data imbalance, and algorithm selection still affect classification performance. Therefore, this study proposes the implementation of several machine learning algorithms for ASD classification using the ASD Toddler Screening Dataset. A comparative approach was used to evaluate the performance of Decision Tree, Random Forest, Logistic Regression, KNN, Naive Bayes, and SVM algorithms. The research stages included data preprocessing, categorical attribute transformation using encoding and one-hot encoding, exploratory data analysis (EDA), data splitting with an 80:20 train-test ratio, and handling imbalanced data using the Synthetic Minority Over-sampling Technique (SMOTE). Model evaluation was conducted using accuracy, precision, recall, and F1-score metrics. The results showed that SVM achieved the best performance with an accuracy of 100%, followed by Logistic Regression (98.58%), Random Forest (95.73%), Decision Tree (93.36%), KNN (92.89%), and Naive Bayes (87.20%). The findings indicate that SVM provides superior performance compared to other machine learning algorithms in classifying ASD in toddlers.
Intelligent Human-Centric Cybersecurity Framework For Adaptive Phishing Mitigation In Higher Education Institutions Lutfiyah Dwi Setia; Imam Riadi; Herman Yuliansyah
MULTITEK INDONESIA Vol 20 No 2 (2026): DECEMBER (ON PROGRESS)
Publisher : Universitas Muhammadiyah Ponorogo

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.24269/mtkind.v20i2.14612

Abstract

Phishing in higher education combines technical exposure with user-level and organizational vulnerabilities, yet the literature often treats these dimensions separately. This study develops an Intelligent Human-Centric Cybersecurity Framework through a structured narrative review and thematic synthesis. Sources were drawn from Scopus, IEEE Xplore, SpringerLink, and ScienceDirect, with the main search window covering 2021–2026. The review screened 268 candidate records and retained 25 sources represented in the final synthesis. A hybrid deductive–inductive coding procedure was used to examine digital literacy, cybersecurity awareness, organizational security culture, behavioral monitoring, and adaptive response while allowing additional themes such as phishing simulation, user compliance, and security fatigue to emerge. The synthesis indicates that human capability and institutional reinforcement become more actionable when they are connected to behavioral evidence and a feedback mechanism. In the proposed framework, “intelligent” refers to the use of monitoring and analytics to support evidence-informed mitigation decisions, whereas “adaptive” refers to the adjustment of training, policy, or controls in response to observed behavior and evolving phishing patterns. The framework is conceptual rather than empirically validated; therefore, controlled phishing simulations and quantitative model testing are proposed as subsequent validation steps.