<![CDATA[Abstrak - Perkembangan teknologi informasi menimbulkan tantangan dalam deteksi anomali log secara real-time. Penelitian ini mengembangkan NetLog, sistem monitoring log dengan pendekatan hybrid yang menggabungkan deep learning dan deteksi berbasis aturan. Sistem menggunakan Autoencoder untuk mempelajari pola log normal dan rule-based detector sebagai fallback. Arsitektur terdiri dari backend FastAPI, frontend React/Next.js, dan modul anomaly detection. Hasil implementasi menunjukkan sistem berhasil mendeteksi 60% serangan simulasi dengan precision 100% dan recall 20%. Evaluasi komprehensif menunjukkan ROC AUC 82% dan PR AUC 86.7%, mengindikasikan kemampuan model yang baik dalam membedakan log normal dan log anomali. Dashboard real-time menampilkan log dengan latensi di bawah 2 detik. Kesimpulannya, pendekatan hybrid pada NetLog terbukti efektif memperluas cakupan deteksi anomali dibandingkan metode tunggal, meskipun masih diperlukan peningkatan sensitivitas deteksi.Kata kunci: Deteksi Anomali; Autoencoder; Deep Learning; Monitoring Log; Sistem Real-time; Abstract - The rapid advancement of information technology poses new challenges in real-time log anomaly detection. This study develops NetLog, a log monitoring system based on a hybrid approach that combines deep learning with rule-based detection. The system employs an Autoencoder to learn normal log patterns and a rule-based detector as a fallback mechanism. The architecture consists of a FastAPI backend, React/Next.js frontend, and an anomaly detection module. The implementation results show that the system successfully detected 60% of simulated attacks with 100% precision and 20% recall. A comprehensive evaluation demonstrates ROC AUC of 82% and PR AUC of 86.7%, indicating a strong ability of the model to distinguish between normal and anomalous logs. The real-time dashboard displays log data with latency below 2 seconds. In conclusion, the hybrid approach of NetLog effectively broadens anomaly detection coverage compared to single-method systems, although improvements in sensitivity are still required.Keywords: Anomaly Detection; Autoencoder; Deep Learning; Log Monitoring; Real-time System;]]>
