Machine learning-based models for self-learning indoor heat warning systems in households

With climate change and global rising temperatures heat health warning systems have become important in accurately predicting heat waves. However, most heat health warning systems rely on the ambient temperature forecast and do not take indoor building conditions into consideration. Moreover, a general heat warning system cannot accurately predict the heat stress conditions in individual buildings. To implement the prediction algorithms the study also proposes a Raspberry Pi based measurement system. Furthermore, to reduce the computational load on Raspberry Pi a Transfer learning technique is implemented from a pre trained Long Short-Term Memory (LSTM) neural network. The results show prediction accuracy of 97% with an RMSE of 0.218 for indoor temperature prediction.