Bi-LSTM Based Real-Time Human Activity Recognition From Smartphone Sensor Data

Human activity recognition (HAR) has emerged as a key field of research in pervasive computing due to its wide range of applications in addressing real-world, human-centric challenges. Three or more sensors that are affixed to various body areas are used in the majority of multi-sensor techniques used in human activity recognition research. But for certain moves, a lot of sensors were unnecessary. With noisy and missing data, a bi-directional LSTM parallel layer can speed up feature extraction. The proposed methodology introduces a Bidirectional Long Short-Term Memory (BiLSTM) model, which incorporates both Linear Discriminant Analysis (LDA) for feature extraction and a univariate filter method for feature selection. This comprehensive approach significantly enhances model performance. Leveraging the UCI HAR dataset, our model achieves an impressive accuracy rate of 97%. This underscores the effectiveness of synergizing deep learning techniques with sophisticated feature engineering strategies. By leveraging the BiLSTM's inherent ability to capture temporal dependencies, alongside LDA's discriminative feature extraction and the filter method's dimensionality reduction, our methodology ensures robust activity recognition across diverse scenarios. Notably, this approach improves accuracy and enhances the model's adaptability to real-world applications, making it well-suited for addressing complex activity recognition challenges efficiently.