Prediction of human activity using machine learning

Abstract

Involving machine learning in recognizing human activities is a widely discussed topic of this era. It has a noticeable growth of interest for implementing a wide range of applications such as health monitoring, indoor movements, navigation and location-based services. The process is implemented gradually through several methods obtaining better accuracy than before. The data of human activities can be collected by wifi module, bioharness or wearable device which can be waist, wrist or thighs mounted. The purpose of our research is predicting human activities by classifying sequences of remotely recorded data of well-defined human movements using responsive sensors. The data are collected by a waist mounted device which contains mobile phone sensors (e.g. accelerometer and gyroscope) for observing human activities of different aged people. The observed data are modeled using machine learning and neural network. Here we have used machine learning algorithms which are Support Vector Machine (SVM), K Nearest Neighbour (KNN), Linear Regression, Logistic Regression, Decision Tree, Naive Bayes Classifier and Random Forest Classifier. Moreover, we have also used artificial recurrent neural network (RNN) architecture- Long Short-Term Memory algorithm and Multi Layer Perceptron (MLP) algorithm. Modeling the data using various algorithms and obtaining results accurately are not convenient, because human motions recorded through wearable sensors have variations and complexity. For overcoming these problems we have used four dimension reduction techniques e.g. Principal Component Analysis (PCA), Singular Value Decomposition (SVD), Linear Discriminant Analysis (LDA) and Independent Component Analysis (ICA) for achieving more accurate activity prediction performance with less complex and faster computations.