OBS-CRN: A Spatio-Temporal Deep Learning Model for Smart Environmental Design Evaluation Using IoT Sensor Data

Abstract

Environmental design plays a critical role in creating spaces that are functional, comfortable, and sustainable. With the integration of Internet of Things (IoT) technology, environmental data is collected in real time, while machine learning (ML) enables intelligent analysis and optimization. Existing evaluation systems often rely on static assessments, limited parameter coverage, and generic models that fail to fully capture the spatio-temporal dependencies in environmental data, resulting in suboptimal design recommendations. Environmental design is critical for developing functional, comfortable, and sustainable environments. This research proposes a complete IoT- ML assessment system that employs the Optimized Beluga Spatial Convolve Recurrence Network (OBS-CRN) to capture spatial correlations across zones and temporal patterns, such as daily and seasonal fluctuations. The framework uses multi-sensor data from an office building (temperature, humidity, CO₂, PM₂.₅, illuminance, sound levels; dataset size: 6,480 + samples) for preprocessing, including missing-value handling and z-score normalization. PMV, PPD, auditory irritation index, daylight autonomy, and NDVI are used to extract features, which are then reduced in dimensionality using PCA. OBS-CRN outperformed baseline models (RNN, GRU, LSTM, CNN, SVR+LGBM+RIDGE, Bi-LSTM, Attention LSTM, CNN-LSTM, and IHHODL-ECP) with an RMSE of 0.25, MAE of 0.25, MSE of 0.10, and R² of 0.990, indicating accurate real-time environmental evaluation. The framework is implemented in Python using TensorFlow, enabling a scalable, real-time, and accurate approach to environmental design evaluation. This system provides an effective and practical tool to improve environmental design through intelligent data analysis and adaptive recommendations.