Energy-Efficient Smart Building Ventilation using Hybrid RNN and Ensemble Regression-Driven Real-Time Environmental Prediction

Abstract

Energy efficiency and indoor environmental quality are critical aspects of modern smart buildings. Traditionally, ventilation systems operated on fixed schedules or simple rule-based mechanisms without considering real-time environmental variations. These conventional systems relied heavily on manual settings or basic sensor thresholds, leading to inefficient energy usage and suboptimal indoor comfort. This research proposes an intelligent ventilation prediction system using a hybrid approach that integrates Machine Learning (ML) and Deep Learning (DL) techniques. The system processes smart home sensor data including temperature, power consumption, occupancy, and environmental conditions. A combination of data preprocessing techniques such as label encoding and feature scaling is applied to prepare the dataset. Multiple regression algorithms, including Gradient Boosting Regressor (GBR) and K-Nearest Neighbors (KNN), are used to model environmental parameters. Additionally, a hybrid Deep Forest Hybrid Regressor (DFHR) model combining a Recurrent Neural Network (RNN) as a feature extractor with a Random Forest Regressor (RFR) is implemented to enhance prediction accuracy. The proposed system predicts key environmental factors such as humidity and light levels in real time, enabling efficient ventilation control. Performance is evaluated using metrics like Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and R² score. The hybrid RNN-based model demonstrates improved prediction performance compared to traditional models. The system is implemented using libraries such as Scikit-learn, TensorFlow, and Tkinter for a user-friendly interface. The research is completed by integrating data processing, model training, evaluation, and real-time prediction into a single interactive application, contributing to smarter and more energy-efficient building management systems