PROJECT TITLE :

A Machine Learning Approach for Fall Detection and Daily Living Activity Recognition

ABSTRACT:

In Western countries, the number of elderly individuals is steadily increasing. The majority of them wish to live freely and are prone to falls. Falls frequently result in serious or even fatal injuries, making them the top cause of death among the elderly. To overcome this issue, it is critical to create reliable fall detection systems. In this context, we create a machine learning framework for detecting falls and recognizing daily life activities. We recognize seven different activities, including falls and activities of daily living, using acceleration and angular velocity data from two public datasets. We extract time- and frequency-domain features from the acceleration and rotational velocity data and feed them into a classification method. We compare the performance of four algorithms for identifying human activities in this research. The artificial neural network (ANN), K-nearest neighbors (KNN), quadratic support vector machine (QSVM), and ensemble bagged tree are examples of these algorithms (EBT). The power spectral density of the acceleration yields new features that increase the classifier's performance. Only the acceleration data are used for activity recognition in the first step. The KNN, ANN, QSVM, and EBT algorithms achieved overall accuracy of 81.2 percent, 87.8%, 93.2 percent, and 94.1 percent, respectively, according to our findings. For the QSVM and EBT algorithms, fall detection accuracy is 97.2 percent and 99.1 percent, respectively, with no false alarms. In a second stage, we extract characteristics from both the acceleration and angular velocity data's autocorrelation function and power spectral density, which enhances classification accuracy. We were able to attain overall accuracy of 85.8%, 91.8 percent, 96.1 percent, and 97.7% for the KNN, ANN, QSVM, and EBT algorithms, respectively, utilizing the proposed features. Both the QSVM and EBT algorithms achieve 100% accuracy in fall detection without any false alarms, which is the greatest possible result.