Contactless Indoor Temperature Sensing via Wi-Fi Channel State Information: A Machine Learning Approach with Real-Time ESP32 Deployment

Indoor temperature monitoring is essential for smart buildings, HVAC optimization, and occupant comfort, yet conventional thermistor-based sensors require dense deployment and maintenance. This paper introduces a novel contactless temperature sensing methodology leveraging Wi-Fi Channel State Information (CSI) modulated by temperature-dependent air refractive index variations. We formulate the CSI-temperature relationship via electromagnetic wave propagation theory and develop a two-stage machine learning pipeline: (1) CSI feature extraction using statistical moments and discrete wavelet transform coefficients across 30 OFDM subcarriers, and (2) regression via Gradient Boosting (XGBoost) trained on 15,000 CSI snapshots spanning 15--35\((^{\circ})\)C in controlled climate chamber experiments. Theoretical analysis establishes sensitivity bounds of \((0.8\times10^{-6})\) refractive index change per \((^{\circ})\)C at 2.4 GHz, translating to observable CSI amplitude variations of 0.3--0.7 dB. Experimental validation across three indoor environments (laboratory, office, residential) achieves mean absolute error (MAE) of 0.68\((^{\circ})\)C and root mean square error (RMSE) of 0.91\((^{\circ})\)C, with inference latency of 42 ms on ESP32 microcontrollers. Comparative analysis demonstrates 23% accuracy improvement over polynomial regression baselines and robustness under multipath-rich conditions. Real-time deployment over 30 days shows 94.2% uptime with drift correction via periodic calibration. The system enables cost-effective (\((<)\)$15/node), maintenance-free ambient monitoring for IoT-enabled smart buildings without physical sensor installation.