Personalized Deep Learning in Human Activity Recognition from Inertial Signals: a Preliminary Study on its Effectiveness

In the recent years there has been a growing interest in techniques able to automatically recognizeactivities performed by people. This field is known as Human Activity recognition (HAR). HAR canbe crucial in monitoring the wellbeing of the people, with special regard to the elder population andthose people affected by degenerative conditions. One of the main challenges concerns thepopulationdiversityproblem, that is, the natural differences between users’ activity patterns, which implies thatexecutions of the same activity performed by different people are different. Previous experiments haveshown that personalization based on similarity between subjects and signals can increase the accuracyof recognition models of human activities obtained by traditional machine learning techniques. In thisarticle, we investigate whether personalization applied to deep learning techniques can lead to moreaccurate models with respect to those obtained both by applying personalization to machine learningmodels, and to traditional deep learning models. In particular, the experiments have been done ontwo public domain datasets and using the AdaBoost classifier and two Convolutional Neural Networks.Preliminary results show that, on average, traditional deep learning outperforms both personalized deeplearning and personalized machine learning techniques.