Affiliations: [a] Shaheed Zulfikar Ali Bhutto Institute of Science and Technology (SZABIST), Islamabad, Pakistan | [b] College of Computer Science and Information Technology, King Faisal University, Al Ahsa, Saudi Arabia
Abstract: With the advent of smart computing, Internet of Things (IoT) and sensor technology, it is now possible to determine the wellness of the elderly living alone in a home equipped with miniaturized sensors. The technology has the potential to seamlessly monitor daily activities performed by humans living under free conditions. Such automatic activity monitoring systems depend on classification techniques, which can effectively interpret activities as normal or abnormal. Although a significant progress has been made in this area, the application of advanced classification techniques are still required to capture various aspects of activities to predict the wellness with high accuracy. In this paper, we propose a wellness classification model called Temporal Weighted Associative Classification (TWAC) by integrating the spatio-temporal contextual information for predicting wellness of the elderly by monitoring their daily usage patterns of household appliances equipped with sensors. The two key components of the model are Data Preparation Module (DPM) and Dynamic Classification Module (DCM). The DPM collects data from sensors and represents it into a format suitable to apply classification by capturing spatio-temporal information of each activity such as time, location, duration and sub-activities, if any. DCM dynamically generates classification rules by identifying correlations among frequent and less frequent activities recorded over a predefined time window. A classification model is then based on these rules to predict normal or abnormal activities. Moreover, the classification model learns from new data and dynamically updates the rules to accommodate the change in daily activities’ pattern, hence, over time the prediction become more accurate. TWAC is tested for its accuracy by comparing it with well-known classifiers such as C4.5, HMM, NB, SVM, CPAR and CBAR. Improved analysis results have been observed as documented in the experimental analysis. Based on a high accuracy, the proposed model will be suitable to develop systems used to forecast the behavior and wellness of the elderly living alone in smart homes.
