The Multimodel Stacking and Ensemble Framework for Human Activity Recognition

Abstract

Human activity recognition (HAR) plays an important role in various domains, such as healthcare, elderly care, sports, gait analysis, and security surveillance. Despite its significance in various fields, attaining a high accuracy remains a formidable challenge. This letter proposes a multimodel stacking and ensemble framework for HAR. The proposed model uses a horizontal stacking approach integrating three different model, namely, ridge regression, LightGBM, and gradient boosting machine (GBM) combined to form a blended model. GBM is also serves as the meta-learner in this configuration. By leveraging this stacking framework, our model significantly enhances the accuracy of HAR. The proposed model achieves remarkable performance in publicly available datasets with accuracy rates of 98% on the HCI-HAR dataset, 99.10% on the WISDM dataset, and 99.20% on the mHealth dataset thereby surpassing existing benchmarks in machine learning. Therefore, the proposed model uses an ensemble stacking model to represent a promising avenue for enhancing HAR and has potential applications in various fields.