<![CDATA[Abstrak - Penelitian ini mengimplementasikan sistem keamanan laci pintar berbasis IoT dan Edge Machine Learning untuk mengatasi kelemahan kunci konvensional dan tingginya false alarm pada sensor getar biasa. Tujuan utama penelitian adalah mengembangkan sistem klasifikasi anomali getaran secara real-time menggunakan algoritma Rule-Based Decision Tree pada mikrokontroler ESP32. Sistem mengintegrasikan sensor MPU-6050, sensor LDR, dan solenoid lock yang dikendalikan melalui dashboard web. Metode klasifikasi menerapkan logika multi-threshold dengan parameter empiris Threshold Force 4.5 m/s² dan Counter Persistence 5 siklus. Berdasarkan pengujian terhadap 30 sampel uji data getaran, sistem menghasilkan akurasi sebesar 100% dalam membedakan pola "senggolan" dan "congkelan". Mekanisme counter terbukti menurunkan tingkat false alarm hingga 0% dibandingkan metode threshold tunggal. Selain itu, sistem menunjukkan responsivitas tinggi dengan rata-rata latensi notifikasi ke dashboard sebesar 2 detik. Penelitian ini berkontribusi pada pengembangan solusi keamanan fisik yang responsif, akurat, dan minim alarm palsu.Kata kunci : Laci Pintar; Pembelajaran Mesin Edge; Deteksi Anomali Getaran; Keamanan IoT; ESP32; Abstract - This research implements a smart drawer security system based on IoT and Edge Machine Learning to overcome the weaknesses of conventional locks and the high rate of false alarms on ordinary vibration sensors. The main objective of the research is to develop a real-time vibration anomaly classification system using the Rule-Based Decision Tree algorithm on an ESP32 microcontroller. The system integrates an MPU-6050 sensor, an LDR sensor, and a solenoid lock controlled through a web dashboard. The classification method applies multi-threshold logic with empirical parameters of Threshold Force 4.5 m/s² and Counter Persistence 5 cycles. Based on testing on 30 vibration data test samples, the system produces 100% accuracy in distinguishing "nudge" and "pry" patterns. The counter mechanism is proven to reduce the false alarm rate to 0% compared to the single threshold method. In addition, the system shows high responsiveness with an average notification latency to the dashboard of 2 seconds. This research contributes to the development of physical security solutions that are responsive, accurate, and have minimal false alarms.Keywords: Smart Drawer;, Edge Machine Learning; Vibration Anomaly Detection; IoT Security; ESP32;]]>
