Analysis of FitBit Data: Detecting Anomalous Activity Patterns

This project uses the following dataset: https://zenodo.org/record/53894#.W2jv3X4nZZ0

It also uses pyvenn to create a 4-circle Venn diagram: https://github.com/tctianchi/pyvenn

Abstract

Fitness tracker research has been heavily focused on by academia in recent years, including efforts to assess fitness, sleep, heart health, general wellbeing, recuperation from medical maladies, and more.  Scholars have used techniques in statistics, machine learning, deep learning, and several other fields to analyze, classify, and predict daily user behavior patterns and outliers in those patterns. In turn, this data has been used to predict and chart medical treatment reactions, encourage weight loss, enforce desired bedtimes, track general fitness, predict if students are studying enough, and give recommendations for current and future behavior to meet fitness goals, just to name a few use cases. This study will focus on finding anomalous behavior patterns within FitBit data, and then finding indicators that predict deviations from behavior baselines. Correlating these activities will be performed by using data mining techniques with Python, on a dataset of 35 users over a 60-day time period in 2016.