Search

The recognition of individual activity has proven its importance in many application areas. Even after the pandemic crisis worldwide, the remote monitoring of human actions and their activities has increased a lot. In such situations, the smart sensors are utilized to capture the details remotely, share, and process intelligently using the computer-assisted machine. This paper introduces the new variant of the machine learning classifier, named Grid Search Based Nonlinear Multi-task Least Squares Twin Support Vector Machine (GMLT-SVM), to recognize human actions using sensory data. The proposed approach uses the grid search for hyperparameter tuning with a Nonlinear kernel for Multi-task classification in a twin support vector machine. This not only increases the efficacy of the model but also reduces the computation time for finding the optimal values of the hyper-parameters. The proposed approach when experimented with the sensory data of Human Action Recognition (HAR) gives an accuracy of 96.8%. The comparative analysis of the proposed method with existing other ML classifiers gives an increase in the accuracy of 94.2±2.6%. The experimental results suggest that the gained results are statistically significant enough to recognize human actions with sensory data. [ABSTRACT FROM AUTHOR]

Recently, the research community has taken great interest in human activity recognition (HAR) due to its wide range of applications in different fields of life, including medicine, security, and gaming. The use of sensory data for HAR systems is most common because the sensory data are collected from a person's wearable device sensors, thus overcoming the privacy issues being faced in data collection through video cameras. Numerous systems have been proposed to recognize some common activities of daily living (ADLs) using different machine learning, image processing, and deep learning techniques. However, the existing techniques are computationally expensive, limited to recognizing short-term activities, or require large datasets for training purposes. Since an ADL is made up of a sequence of smaller actions, recognizing them directly from raw sensory data is challenging. In this paper, we present a computationally efficient two-level hierarchical framework for recognizing long-term (composite) activities, which does not require a very large dataset for training purposes. First, the short-term (atomic) activities are recognized from raw sensory data, and the probabilistic atomic score of each atomic activity is calculated relative to the composite activities. In the second step, the optimal features are selected based on atomic scores for each composite activity and passed to the two classification algorithms: random forest (RF) and support vector machine (SVM) due to their well-documented effectiveness for human activity recognition. The proposed method was evaluated on the publicly available CogAge dataset that contains 890 instances of 7 composite and 9700 instances of 61 atomic activities. The data were collected from eight sensors of three wearable devices: a smartphone, a smartwatch, and smart glasses. The proposed method achieved the accuracy of 96.61% and 94.1% by random forest and SVM classifiers, respectively, which shows a remarkable increase in the classification accuracy of existing HAR systems for this dataset. [ABSTRACT FROM AUTHOR]

Lately, the continuous temporal data from motion sensors in wearable devices has been great interest for the research community due to its demand for analyzing human activities in several applications e.g. healthcare, sports, and surveillance. Numerous solo deep learning models have been proposed in the literature to extract an adequate feature representation from temporal sensory data, however, they are limited to encode only one aspect of the data, and not sufficient to capture the in-depth association between the patterns. This paper presents an ensemble deep model to encode the complex patterns and dependencies in temporal data. An ensemble model refers to the merging or blending of separate deep models, aiming to utilize their unique strengths and abilities to construct a more resilient and effective solution. The contributions of this paper are two folds: First, a systematic evaluation of five different ensemble models is presented to encode the raw sensory data. Secondly, this study proposed a lightweight, multi-layer, and hybrid LSTM-GRU model to recognize the human activities using time series sensory data. Specifically, the raw sensory data is fed into a two-layer LSTM followed by GRU layers. To avoid over-fitting and accelerate convergence, we employed dropout and batch normalization layers. An extensive evaluation of all the five ensemble models is carried out on two large benchmark datasets, namely: UCI-HAR and WISDM. The proposed LSTM-GRU architecture achieved the recognition accuracy of 99.06% and 96.61% on the WISDM and UCI-HAR, respectively. The performance comparison with existing state-of-the-art techniques confirms the superiority and efficacy of the proposed algorithm. • A comprehensive literature review of existing HAR techniques is conducted. • A systematic evaluation of five different ensemble models is presented. • The optimal network configuration is investigated to improve accuracy and efficiency. • The performance of proposed model is evaluated on two large and diverse datasets. • A comparison of various architectures is presented to analyze the sensory data. [ABSTRACT FROM AUTHOR]