Interpretability in machine learning for IAQ and HVAC optimisation: A response to Oka et al

Abstract

Godasiaei et al. employed advanced deep learning models, including– GRUs, RNNs, LSTMs, and CN – to capture temporal and spatial patterns in air pollution data. The reported methodology addresses four critical challenges: (1) Model Architecture Optimization through systematic weight/bias adjustment, hyperparameter tuning, and hidden layer configuration; (2) Bias Mitigation using G-DeepSHAP and CNN-assisted visualization; (3) Rigorous Validation via k-fold cross-validation and sensitivity analysis; and (4) Practical Implementation bridging theoretical constructs with real-world indoor air quality (IAQ) management. By combining machine learning with sensitivity analysis – supported by empirical validation and systematic model refinement – this research overcomes key limitations of traditional air pollution analysis methods.