The goal of the project is to create a Neural Network for multi class classification problem to recognize the human activities with the help of statistical features extracted from sensor mounted on various parts of the body. The open Motion Sense data set is downloaded from online.
Multi Layer Perceptron was built for the problem in which various hyper parameters are selected and the best fit is chosen based on the grid search. Cross Entropy was used as the loss function whereas adam optimizer is used to optimize the loss. The network achieved 94% of accuracy with 0.94 F1 score which is considered to be good considering the simplicity of the machine.