Efficient Human Activity Recognition using PCA Dimensionality Reduction and GWO-Enhanced LSTM

Abstract

Human activity recognition (HAR) is a critical task in various applications such as healthcare, smart homes, and security systems. The high dimensionality of the input data and the challenge of optimizing model parameters can hinder the development of accurate and efficient HAR systems. This research presents a novel method that combines Principal Component Analysis (PCA) for dimensionality reduction with Long Short-Term Memory (LSTM) networks. The LSTM network is optimized using the Grey Wolf Optimization (GWO) algorithm to enhance classification performance. Initially, PCA is employed to reduce the dimensionality of the input feature space, significantly reducing the number of features from 561 to 196 while preserving more than 99% of the original data variance. This step improves computational efficiency and reduces the risk of overfitting. The GWO algorithm is then used to fine-tune the LSTM network's hyperparameters, including the number of hidden units, learning rate, drop frequency, and batch size. This optimization ensures that the LSTM network effectively captures complex temporal dependencies in the activity data. The proposed method was rigorously tested and achieved a remarkable accuracy of 98.95%, demonstrating its robustness and efficacy in human activity recognition tasks. The integration of PCA and GWO improves the model’s performance and enhances its generalization capability to new, unseen data. This approach offers a powerful and efficient solution for HAR, addressing fundamental challenges and paving the way for future advancements in the field.