Deep Learning Model for Human Activity Recognition and Prediction in Smart Homes
Cheng Wang, Zhihao Peng · 2020
To solve the limitation problem of traditional human activity recognition (HAR) tasks which use features extracted manually and some shallow machine learning models, a novel multi-task layer neural network (LSTM) model is proposed based on the ability of deep neural network to automatically extract features in a smart home, combined with the recent successful recurrent neural network and convolution neural network. Prediction techniques based on LSTM neural networks in smart home environments have been used to predicting the next activity as well as on the task of predicting the timestamp of the next event as well. The performance of the model is evaluated on the real dataset. Experimental results show that the LSTM neural networks outperform the other approaches for the prediction of the direct next event, meanwhile, the application of multi-task learning by jointly predicting the next activity and the timestamp of the next event outperforms separate LSTM models for both tasks separately.