BST 593 Applied Project

---

📘 Introduction

This study presents an unsupervised machine learning approach for anomaly detection in wearable-device data, using inter-model agreement to identify irregular patterns.

This repository contains the full anomaly detection pipeline applied to the Kaggle Health and Fitness dataset.
The dataset includes 3,000 participants and more than 650,000 rows of time-series measurements.

---

📦 Required Libraries

Python

• pandas — dataframe manipulation
• numpy
• pyarrow — required for reading/writing Parquet files
• scikit-learn
• torch (PyTorch)
• torchvision
• torchaudio
• notebook / ipykernel

R

• tidyr
• dplyr
• ggplot2
• arrow
• glmmTMB
• xtable
• mltools
• tidyverse
• irr
• purrr

---

📂 Repository Structure

├── 01_Introduction
│   ├── Fitness_data_pipeline.R
│   └── Introduction.rmd
├── 02_DataAnalysis
│   ├── AppliedProject.ipynb
│   └── src
│       ├── algorithms.py
│       ├── dataset.py
│       └── preprocessing.py
├── 03_results
│   ├── BST593-lme.Rmd
│   └── utils
│       └── analysis.R
├── data
│   └── pipeline
│       ├── autoencoder_running.parquet
│       ├── autoencoder_swimming.parquet
│       ├── DATA.md
│       ├── fitness.parquet
│       ├── fitness.rds
│       ├── health_fitness_dataset.csv
│       ├── kMeans_running.parquet
│       ├── kMeans_swimming.parquet
│       ├── oneClassSVM_running.parquet
│       └── oneClassSVM_swimming.parquet
├── figure
│   ├── activity-pie-chart-1.pdf
│   ├── avg-hr-boxplot-1.pdf
│   ├── diastolic-bp-boxplot-1.pdf
│   ├── health-condition-bar-1.pdf
│   ├── kmeans_anomaly_Running.png
│   ├── kmeans_anomaly_Swimming.png
│   ├── kmeans_parameter_Running.png
│   ├── kmeans_parameter_Swimming.png
│   ├── oneClassSVM_Running.png
│   ├── oneClassSVM_Swimming.png
│   ├── resting-hr-boxplot-1.pdf
│   └── systolic-bp-boxplot-1.pdf
└── README.md

---

📑 Folder Descriptions

01_Introduction

Includes data preprocessing and summary statistics, such as missing-value exploration, descriptive visualizations, and hypothesis testing.

02_DataAnalysis

Contains the main analysis workflow. Machine-learning–based anomaly detection is implemented using:

• k-means clustering
• one-class SVM
• autoencoders

03_results

Includes the linear mixed-effects logistic regression models used to analyze covariates in the activity-tracking data.

data/pipeline

Stores processed outputs from each step of the workflow, including:

• CSV
• RDS
• Parquet files

A detailed description of each dataset is available in
➡️ DATA.md

figure

Contains all visualizations, including:

• dataset summaries
• anomaly detection results
• model parameter optimization plots

---