الفهرس 
AbstractOnly 14 pages are availabe for public view
 |  | from | 121 |  |  |
| | | from | 121 | | |
Abstract
Human Activity Recognition (HAR) has become important as there are many other important applications are depending on it. For example, it can be used as basis for customized medicine and to assistant living of the elderly as they need more home care. HAR can help older people by providing a continuous remote monitoring. This can help with predicating falls and also help in predicating some diseases and illnesses in their early stages. In recent years, using deep learning techniques has grown steadily in different fields affecting the day-to-day decisions of individuals. Smartphone and wearable devices can provide huge amount of data collected that could be used for building such models. Combining deep learning and Internet of Things (IoT) can helps in increasing the overall performance of HAR systems. This thesis proposed and developed two intelligent models for HAR based on deep learning. The first proposed model is based on capsule neural network to build a new model (1D-HARCapsN). This proposed model consists of four layers: convolution layer, primary capsule layer, activity capsules flat layer and output layer. It is normalized using the random-SMOTE algorithm to handle the imbalanced behavior of the dataset. The results of the practical experiments that conducted on WISDM dataset indicated that the proposed this model enhanced the performance with accuracy 98.67%, precision 98.66%, recall 98.67%, and F1- measure 0.987. This means that this model shows a major performance enhancement compared to the other compared works. The second proposed model, Ensemble of Models for Human Activity Recognition (EMHAR), is built based on Ensemble Classifiers of the deep learning models (ANN, CNN, RNN, and LSTM) that are deployed on the Edge node. This ensemble uses majority vote from these models and produces a final predication as the predication of most of those models. The results of practical experimentations showed that model achieved accuracy up to 99.2% and 97.4% based on our collected physical activities dataset that consists of eight different activities and UCI-HAR dataset, respectively. The experimentations result also showed that this model outperformed the other compared models regarding the accuracy based on UCI-HAR dataset.