Supervised Machine Learning Based Behavioral Pattern Recognition for Autonomous Non-Invasive Smart Switch Actuation Using a Fingerbot System

The proliferation of smart home technologies has intensified demand for adaptive, low-cost automation solutions deployable without altering existing electrical infrastructure. This paper presents an AI-driven Fingerbot, a non-invasive electromechanical actuator that physically operates conventional wall switches through a micro servo motor governed by an ESP32-C3 microcontroller. Unlike solutions that require electrical rewiring or hardware replacement, the device mounts externally on standard switches and automates their operation through learned behavioral patterns. Three supervised classifiers; Logistic Regression, Decision Tree, and Random Forest were trained on 1,440 time-stamped user interaction records capturing hour, minute, day of week, and weekend indicator features to predict binary switch states. The Decision Tree classifier achieved 99.7% accuracy, 99.6% precision, and 99.4% recall. Random Forest recorded 97.1% accuracy with the highest precision of 99.8%, while Logistic Regression served as a linear baseline at 87.9% accuracy. Response time evaluation across three control modes showed that physical switch interaction averaged 311.67 ms (maximum 400 ms), Wi-Fi virtual control averaged 2,597.33 ms (maximum 3,750 ms), and mobile hotspot control averaged 3,257.67 ms (maximum 5,750 ms). The embedded AI activation logic commenced switch-on at approximately between 18:35 to 18:45 and enforced mandatory deactivation at 23:59 to minimize unnecessary energy consumption. Monthly energy monitoring of a 25 W test load yielded figures ranging from 121.80 kWh to 143.53 kWh across three consecutive months. The results confirm that integrating lightweight machine learning with low-cost actuator hardware constitutes a practical and cost-effective pathway for retrofit smart home automation.