Paper Title
MACHINE LEARNING BASED HUMAN ACTIVITY DETECTION

Abstract
Abstract - Human Activity Recognition (HAR) aims to identifiy human activities based on sensor estimations, as well as to recognise accurate and efficient human behaviour represents as a challenging field of research in computer vision. To overcome the challenges, the two basic models: Convolutional Neural Network (CNN) and Deep Learning Long Short-Term Memory (LSTM) had the accurate results for all users, with 91.77% and 92.43%, respectively. Convolutional Neural Networks (CNNs) have emerged as a useful category of systems for issues involving image recognition or computer vision. We investigate several methods for increasing a CNN's connection in the time domain to take benefits from local spatio-temporal data, and recommend a multi-resolution, foveal structure as a potential method to quicken training. We propose an experimental and improved approach that combines improved hand-crafted features with neural network architecture that outperform powerful methods while applying the same standardized score to different datasets. Finally, we offer a variety of analysis- related suggestions for researchers. This survey report is a valuable resource for people interested in future research on human activity recognition. Keywords - Convolution Neural Network, Deep Learning, Human Activity Detection, Accelerometer Data, Long-Short Term Memory, Wireless Sensor Data Mining