Optimizing Edge Computing for Activity Recognition: A Bidirectional LSTM Approach on the PAMAP2 Dataset

Abstract

This study investigates the application of a Bidirectional Long Short-Term Memory (BiLSTM) model for Human Activity Recognition (HAR) using the PAMAP2 dataset. The aim was to enhance the accuracy and efficiency of recognizing daily activities captured by wearable sensors. The proposed BiLSTM-based model achieved outstanding performance, with 98.75% training accuracy and 99.27% validation accuracy. It also demonstrated high precision, recall, and F1 scores (all 0.99). Comparative analysis with state-of-the-art models, including Deep-HAR and CNN-BiLSTM-BiGRU, revealed that the proposed BiLSTM model surpassed their performance. These results highlight the potential of the proposed approach for real-time HAR applications in edge computing, particularly where accurate and efficient activity recognition is crucial.