Clustering-Anomaly-Detection-Projects

Applying unsupervised clustering and anomaly detection techniques to health-related datasets (heart disease prediction and wearable sensor data).

Projects

1. Heart Disease Prediction (K-Means Clustering)

   • Goal: To group patients into clusters based on medical features (age, sex, cholesterol, max heart rate, etc.) without using pre-defined labels ('heart disease' vs. 'no heart disease').
   • Algorithm: K-Means Clustering
   • Data: Heart disease dataset (data/heart-disease.csv)
   • Process:
     • Load and preprocess data (handle categorical features).
     • Standardize features for equitable distance calculations.
     • Apply K-Means with k=2 clusters.
     • Predict cluster assignments for the data.
     • (Optional) Evaluate cluster assignments against true labels (for learning purposes).
   • File: heart_disease_Kmeans.py

2. Wearable Time-Series Anomaly Detection (KNN)

   • Goal: To detect anomalous (unusually high) overnight resting heart rates (RHR) from smartwatch data, which could indicate early signs of infection or other physiological stress.
   • Algorithm: K-Nearest Neighbors (KNN) for Outlier Detection (via PyOD library)
   • Data: Wearable heart rate data (data/P100300/Orig_Fitbit_HR.csv)
   • Process:
     • Load time-series data.
     • Resample data to get daily median overnight RHR.
     • Apply KNN outlier detection to identify points significantly distant from their neighbors.
     • Visualize the time series, highlighting the detected anomalies. [
   • File: wearable_anomaly_knn.py

Requirements

• Python 3.x
• pandas
• scikit-learn
• matplotlib
• pyod (for the wearable data project)
• numpy

Install requirements using pip:

pip install pandas scikit-learn matplotlib pyod numpy