Revolutionizing Big Data Assessment for Human Activity Recognition with Metapath Context and Bi-directional Cascade Networks

Abstract

Abstract Human activity recognition focuses on automatically detecting and classifying what individuals are doing, utilizing data gathered from sensors or video feeds. Challenges include interpreting complex activities, handling imbalanced data classes, ensuring privacy in video-based methods, and achieving efficiency with limited computational resources. To solve these issues, this research develops a novel model named the Metapath Contexted Convoluted Bi-directional Cascade Network (MCCBCN), which leverages meta paths to enhance accuracy and robustness in capturing contextual information for human activity recognition. By leveraging both convolutional and bi-directional cascade architectures, MCCBCN makes a significant leap in the field by effectively capturing temporal dependencies. With the addition of the Enhanced Football Team Training Optimization Algorithm (EFTTOA), training efficiency gets a boost, allowing the model to better understand the complex temporal and contextual nuances in human activity data. Additionally, utilizing Large-scale Synthetic Minority Over-Sampling Technique (LSMOST) for dataset expansion ensures a more comprehensive and balanced representation of minority classes, mitigating biases from imbalanced class distributions and improving model generalizability. The proposed method achieved impressive evaluation metrics with an accuracy of 99.50%, F1-score of 99.9%, precision of 99.42%, and recall of 99.3%, outperforming existing techniques. Also, the proposed MCCBCN time complexity is 10.2 seconds. The proposed MCCBCN model effectively addresses key challenges in human activity recognition by enhancing contextual understanding, improving training efficiency, and ensuring robust performance across imbalanced datasets.