Chaos-Informed 1D CNNs for Gait Dynamics Classification

This paper presents a chaotic-adaptive one-dimensional convolutional neural network (1D CNN) architecture for classifying gait dynamics. The novel approach uses a nonlinear dynamic modeling concept to learn subtle deviations in gait patterns, which enables the identification of both normal and constrained movements. A computationally efficient processing framework, consisting of several layers including two convolutional layers, two max-pooling layers, dropout layers, global max-pooling, and a fully connected classification l ayer, is demonstrated, which achieved a test accuracy of 84.87% across validation tests. The model maintained strong generalisation capability and effectively detected variations caused by external physical restrictions without overfitting. This model has the potential to enable the development of efficient, adaptive classifiers for applications in gait analysis, rehabilitation monitoring, and early mobility disorder detection.