Room Occupancy Detection Based on Random Forest with Timestamp Features and ANOVA Feature Selection Method

To improve energy efficiency, understanding occupant behavior is crucial for adaptive temperature control and optimal electronic device usage. Our study introduces a room occupancy detection system using machine learning and Internetof-Things sensors to predict occupant behavior patterns. Initially, indoor IoT sensor devices are installed to observe occupant behavior, and datasets are generated from sensor data, including temperature, humidity, light, and CO2 levels, in both occupied and vacant rooms. The collected dataset undergoes analysis through a machine learning-based model designed to classify room occupancy. First, the timestamp features, extracted from date-time data, such as time of day and part of the day, are extracted. ANOVA feature selection is applied to identify five crucial features. Ultimately, the random forest model is employed to classify room occupancy based on the selected features. Results indicate that our proposed model significantly outperforms other models—achieving improvements of up to 99.713%, 99.467%, 99.676%, 99.676%, and 99.571% in accuracy, precision, recall, specificity, and F1-score, respectively. The trained model holds potential for integration into web-based systems for real-time applications. This predictive model is poised to contribute to the optimization of electronic device efficiency within a room or building by continuously monitoring real-time room conditions.