The essential human gait parameters are briefly reviewed, followed by a detailed review of the state of the art in deep learning for the human gait analysis. The modalities for capturing the gait data are grouped according to the sensing technology: video sequences, wearable sensors, and floor sensors, as well as the publicly available datasets. The established artificial neural network architectures for deep learning are reviewed for each group, and their performance are compared with particular emphasis on the spatiotemporal character of gait data and the motivation for multi-sensor, multi-modality fusion. It is shown that by most of the essential metrics, deep learning convolutional neural networks typically outperform shallow learning models. In the light of the discussed character of gait data, this is attributed to the possibility to extract the gait features automatically in deep learning as opposed to the shallow learning from the handcrafted gait features.

Deep learning models are used to process and fuse raw data of gait-induced ground reaction force (GRF) for Parkinson’s disease (PD) patients and healthy subjects with the aim to categorize PD severity. This is achieved by learning automatically, end-to-end, the spatiotemporal GRF signals, resulting in an effective PD severity classification with mean performance F1-score of 95.5% and F1-score standard errors of 0.28%. Layer-wise relevance propagation (LRP) is used to interpret the models’ output and provide insight into which features in the spatiotemporal gait GRF signals are most significant for the models’ predictions. This allows their assignment to gait events, implying that while for the classification of healthy gait the heel strike and body balance are the most indicative gait elements, foot landing and body balance are those most affected in advanced stages of PD. The proposed models are resilient to noise and are computationally efficient for processing and classification of large longitudinal GRF signal datasets, therefore they can be useful for detecting deterioration in the postural balance and rating PD severity.

The complexity of sensor signal patterns in healthcare, coupled with the variability of physiological data present significant challenges in developing reliable diagnostic and monitoring. While machine learning (ML) has greatly advanced sensor-based health analysis, its decision-making processes often lack transparency, raising concerns about reliability and clinical adoption. This review explores the role of Explainable Artificial Intelligence (XAI) in enhancing the interpretability of ML models for sensor signal analysis, particularly in applications such as pressure sensors, wearable inertial sensors, imaging sensors (MRI, CT, X-ray), ECG, EEG, and wearable health tracking. A systematic literature review was conducted across multiple databases to identify studies applying XAI techniques to sensor-based health monitoring. The review categorizes nine trending XAI methods used to interpret ML-driven analyses of biosignals, evaluating their advantages and limitations in different healthcare scenarios. The findings emphasize the importance of transparency in ML-driven sensor analysis, which is critical for building trust and real-time clinical decisionmaking and wearable healthcare applications. Despite its potential, XAI for sensor signals faces challenges related to model scalability, real-time processing, and clinician interpretability. The review identifies key research gaps in integrating XAI into sensor-based healthcare systems, emphasizing the need for robust validation methods and user-friendly explanations. Ultimately, XAI offers a promising path toward revolutionizing sensor-driven health monitoring, though further advancements are necessary to fully integrate explainability into real-world clinical and assistive applications.

The recognition of gait pattern variation is of high importance to various industrial and commercial applications, including security, sport, virtual reality, gaming, robotics, medical rehabilitation, mental illness diagnosis, space exploration, and others. The purpose of this paper is to study the nature of gait variability in more detail, by identifying gait intervals responsible for gait pattern variations in individuals, as well as between individuals, using cognitive demanding tasks. This work uses deep learning methods for sensor fusion of 116 plastic optical fiber (POF) distributed sensors for gait recognition. The floor sensor system captures spatiotemporal samples due to varying ground reaction force (GRF) in multiples of up to 4 uninterrupted steps on a continuous 2×1 m area. We demonstrate classifications of gait signatures, achieving up to 100% F1-score with Convolutional Neural Networks (CNN), in the context of gait recognition of 21 subjects, with imposters and clients. Classifications under cognitive load, induced by 4 different dual tasks, manifested lower F1-scores. Layer-Wise Relevance Propagation (LRP) methods are employed to decompose a trained neural network prediction to relevant standard events in the gait cycle, by generating a “heat map” over the input used for classification. This allows valuable insight into which parts of the gait spatiotemporal signal have the heaviest influence on the gait classification and consequently, which gait events, such as heel strike or toe-off, are mostly affected by cognitive load.

Introduction: Gait analysis, an expanding research area, employs non-invasive sensors and machine learning techniques for a range of applications. In this study, we investigate the impact of cognitive decline conditions on gait performance, drawing connections between gait deterioration in Parkinson's Disease (PD) and healthy individuals dual tasking.Methods: We employ Explainable Artificial Intelligence (XAI) specifically Layer-Wise Relevance Propagation (LRP), in conjunction with Convolutional Neural Networks (CNN) to interpret the intricate patterns in gait dynamics influenced by cognitive loads.Results: We achieved classification accuracies of 98% F1 scores for PD dataset and 95.5% F1 scores for the combined PD dataset. Furthermore, we explore the significance of cognitive load in healthy gait analysis, resulting in robust classification accuracies of 90% ± 10% F1 scores for subject cognitive load verification. Our findings reveal significant alterations in gait parameters under cognitive decline conditions, highlighting the distinctive patterns associated with PD-related gait impairment and those induced by multitasking in healthy subjects. Through advanced XAI techniques (LRP), we decipher the underlying features contributing to gait changes, providing insights into specific aspects affected by cognitive decline.Discussion: Our study establishes a novel perspective on gait analysis, demonstrating the applicability of XAI in elucidating the shared characteristics of gait disturbances in PD and dual-task scenarios in healthy individuals. The interpretability offered by XAI enhances our ability to discern subtle variations in gait patterns, contributing to a more nuanced comprehension of the factors influencing gait dynamics in PD and dual-task conditions, emphasizing the role of XAI in unraveling the intricacies of gait control.