Time Series Classification Using Convolutional Kernel and Adaptive Dynamic Thresholding

Time series classification is a challenging and essential task in many domains such as finance, healthcare, and industrial systems. Traditional methods often struggle with the complexity of multivariate time-series data and the need for effective feature selection. To address these issues, we propose a novel approach incorporating the Metropolis-Hastings algorithm within a Markov Chain Monte Carlo (MCMC) framework for optimized feature selection. Within the inner workings of the Metropolis-Hastings algorithm, we utilize convolutional kernels and randomized threshold exceedance rates for robust feature extraction. To further enrich the diversity and efficacy of the feature set, we seamlessly integrate Locality-Sensitive Hashing (LSH) techniques into the Metropolis-Hastings framework. As a result, our method not only sets a new standard for efficiency in time-series classification but also surpasses existing state-of-the-art methods across a diverse range of datasets.