Cross-Validated Ensemble ARDenseNet for Human Activity Recognition

Due to advances in hardware embedding technologies, human activity identification utilizing sensors is a challenging study area. These have difficulty identifying actions that are significantly connected, such as a forward fall and a fall from a chair. A novel ensemble deep learning network has been proposed that incorporates training at several levels of abstraction. The UniMiB HAR dataset's fall signals are taken into account when evaluating the proposed network. The Synthetic minority oversampling technique (SMOTE) also helps to balance out the data size inconsistencies. The proposed deep learning network has been shown to have an overall accuracy of 96.3%.