Predicting of Sleep Behaviour in Smart Homes Based on Multi-residents Using Machine Learning Techniques

Abstract

Sleep is a vital part of our daily lives; it plays a significant role in the overall health and welfare of a person. Smart Homes plays a vital role in recognizing human activities performed at home and making it a better place to live. This study focused on assessing sleep behaviour from the ARAS dataset generated from multi-residents in Smart Homes to predict potential sleep behaviour of residents in relation to other activities using different Machine Learning Techniques. In this regard, we used Logistic regression (RL), Linear Discriminant Analysis (LDA), K-Nearest Neighbour (KNN), Naive Bayes (NB), Classification and Regression Trees (CRT) and Support Vector Machine (SVM) to learn and analyze sleep behaviour in relation to other activities of which impacts to the health issues of the residents. The experimental results show that Support Vector Machine (SVM) outperformed in both House A and B in predicting potential sleep activity compared to other algorithms. In House A, Resident1 obtained an accuracy of 90%, and Resident2 obtained an accuracy of 80%, while in House B, Resi-dent1 achieved an accuracy of 100%, and Resident2 obtained an accuracy of 90%. Hence, these results show that resident1 for both House A and B spend enough time for sleeping activity compared to other activities; in contrast, resident2 in both House A and B spend less time in sleeping activity in relation to other activities of which this may affect their health due to sleeping disorder.

Computational methods for predicting human behaviour in smart environments

“…Regression, particularly linear regression like that found in classifiers such as an SVM, can be used for forecasting the next event in a sequence of known linear historical data and applying a best fit to determine the most probable class or category the next event. This is unlikely to scale well for data with many different behavioural events, or correctly predict outlying events in the data [17].…”

“…We can see in Vectors are the simplest of all these data structures. We can imagine a sequence of events represented as sensor values (27,14,1,903,43,17) and a vector is simply a 1-dimensional representation of some sequence of behavioural events. To enrich this data we may want to create a 2-dimensional representation of the data.…”

“…Prediction algorithms or methods fall into three categories: Machine Learning (ML), Probabilistic Graphical Models (PGMs), and Statistical and Trend Analysis [9]. Machine learning algorithm variants include Reinforcement Learning (RL) [10] [11], Neural Networks (NN) [12] [13] [14] [15], Recurrent Neural Networks (RNN) [16], Long Short Term Memory (LSTM) [16] [17], and Support Vector Machines (SVM) [17]. Probabilistic Graphical Models include SPEED algorithm variants (sequence prediction via enhanced episode discovery) [18] [19] [20] and CRAFFT algorithm variants (current activity and features to predict next features / Bayesian network) [21].…”