Comparative analysis of multilayer backpropagation and multi-channel deep convolutional neural network for human activity recognition

Time Series Classification (TSC), is the issue of anticipating class marks of time arrangement information in the area of data mining and machine learning. In any case, regardless it stays testing and misses the mark concerning order exactness and proficiency. To address these issues, a novel Multi-Channel Deep Convolutional Neural Network (MCDCNN) is proposed, which consolidates extraction and characterization in a solitary system. The assessment and execution examination of MCDCNN strategy and the current Multilayer Backpropagation (MLBP) calculation is completed on a PAMAP2 (physical action checking) dataset. The dataset contains unmistakable information gathered from nine human subjects performing six physical exercises like sitting, standing, running, strolling, upstairs and ground floor. Further, the dataset is separated into 734000 preparing tests and 314572 test tests. The starter exploratory investigation of PAMAP2 is finished utilizing WEKA apparatus on the order calculations in particular J48, Random Forest and Naives Bayes. Further, the MLBP and MCDCNN calculations are assessed on PAMAP2 dataset utilizing keras successive model with tensor flow as backend. The novelty of the proposed research lies in classifying a huge dataset (PAMAP2) on three different aspects namely existing tool, machine learning algorithm, and deep learning algorithm. In addition, the results obtained are validated using appropriate quantitative metrics.