Abstract

Lights and fans in many classrooms keep running after the last student has left, or while daylight is already enough, which raises electricity bills for no benefit. Earlier smart-building studies usually either measure energy or switch appliances, and few of them join occupancy sensing, a learned decision model and an inexpensive working prototype. This paper describes an AIoT prototype for a laboratory-scale classroom. An ESP32 board reads IR seat sensors, an LDR and a DHT22 sensor, and a Random Forest classifier, trained on 1,050 cleaned records with an 80:20 split, predicts whether the light and the fan should be on. The model reached 96.2% accuracy on the held-out data, with precision and recall between 94.3% and 95.8%, and fixed thresholds (light level ≤ 250, temperature ≥ 28 °C) keep the behaviour predictable. The circuit was simulated on the Velxio platform, the prototype was demonstrated at the project exhibition on 17 October 2026, and its approximate cost is Rs. 1,960. Comparing always-on operation with controlled operation suggests savings of roughly 40–60%, but this is an estimate that still needs metered confirmation.

Keywords

How to Cite This Article

Rights