Abstract
Across the globe, Smartphones enjoy a high ownership rate. A person’s day-to-day activities are incomplete without the use of smartphones. These sensor-rich devices are often repurposed to collect various forms of data from the user in the domains of 1) app usage, 2) overall phone activity, 3) location, 4) day and night-time activity, 5) communication, 6) social media behavior and so on. This information collected from sensor and log data taken from smartphones can be used to create predictive measures extending to an individual’s Big Five personality traits and examine them. Human activity recognition (HAR) as a field has always tried to improve its performance in the aspects of feature extraction and improve the accuracy of recognition and prediction by performing various machine learning techniques on the data received from accelerometer and gyroscope sensors and creating convolutional neural networks to recognize the multiple activities such as walking, jogging, walking up or down the stairs, etc. Although HAR is a recognized research field, there have been significant challenges such as a) different movements for different individuals, b) limited training data, and c) complexity of performed activities. The idea of this project is to create a HAR model with the help of Machine Learning and Deep Learning, which will be more efficient in terms of performance and accuracy than the conventional models and try to overcome the challenges. That will be applied to the Smartphone data to predict a smartphone user’s activity at one point of the day. The idea is to also get an analysis of the behavioral traits of the user in terms of online expenditure and predict the date of the next purchase. That will be done to send personalized information such as offers and announcements to the user catered according to their needs to enhance their personal experience.
Keywords: Human Activity Recognition, Convolutional Neural Network, Smartphones, Sensor, Machine Learning, Deep Learning.