Evaluation of Fall Detection Using Machine Learning

Abstract

Fall is considered as one of the greatest risks and fundamental problems in health-care for older adults living alone at home. The number of older adults living alone in their own homes is increasing worldwide due to the high expense of health care services. Therefore, it is important to develop a fall detection approach. In this thesis, an intelligent approach is proposed to detect the fall from activities of daily life. This approach employs machine learning techniques, these techniques are linear discriminant analysis, support vector machine, Naïve Bayes, and decision tree. The proposed approach is tested and evaluated based on a publicly available University of Rzeszow Fall Detection (URFD) dataset. This dataset contains two different features, namely depth maps from a Kinect sensor and tri-axial data from an accelerometer. The existing systems suffering from a lack of high accuracy. Therefore, there are high false alarms. Several problems are occurring in systems because of the fusion of data from multi-sensors for FD. Data fusion methods blend data from more than the individual sensor and complementary information from related databases It involves arguments from the specific measurement of data, data transmission to reliable data analysis. The thesis makes the following four contributions: Building a proposed approach utilized a vision-based integrated with a wearable sensor based . Reducing data dimension by using feature selection to reach the highest rate of accuracy and lowest time. Applied four of ML techniques, are LDA, SVM, NB, and DT, then compared the results between these techniques. The proposed approach obtained the highest accuracy compared with the previous studies used the same URFD. For evaluation, several assessment measures are computed. The best technique performance in terms of accuracy and time of implementation is the support vector machine. These evaluation measures demonstrate the effectiveness of the proposed approach when compared with the earlier studies