PROJECT 01 · Full-stack machine learning · Wearable analytics

Proactive Health Insight System
Finding meaningful signals in wearable data

A privacy-preserving analytics prototype for simulated wearable time series. PHIS combines anomaly detection, an API and database layer, and a live dashboard to make unusual patterns easier to inspect.

PHIS / WEARABLE SIGNALS

410K+

Bhanu Teja Malineni01 / 04

RoleProject developer

PeriodProject work

ContextPersonal / academic project

FocusFull-stack machine learning

The idea

Wearables generate continuous streams of measurements, but large volumes can make it hard to spot unusual patterns. I designed PHIS as a full-stack machine learning pipeline that analyzes simulated wearable data and presents results in a visual dashboard.

How it works

The system processed more than 410,000 simulated time-series records across 15+ device profiles. It used Z-score analysis and Isolation Forest for anomaly detection, a Flask REST API backed by PostgreSQL, and a Streamlit and Plotly dashboard for real-time visualization.

01IngestProcess simulated time-series records across multiple device profiles.

02DetectApply Z-score and Isolation Forest methods to identify anomalies.

03ExploreServe results through a Flask API and visualize them in Streamlit + Plotly.

Reported outcomes

410K+Simulated time-series records processed.

15+Simulated device profiles represented.

2 methodsZ-score and Isolation Forest anomaly detection.

Full stackFlask REST API, PostgreSQL, Streamlit, and Plotly.

The records and device profiles are simulated. The project is an analytics prototype, not a clinical diagnostic system.

Tools & methods

PythonScikit-learnIsolation ForestFlask REST APIPostgreSQLStreamlitPlotlyTime-series analytics

More about the project

The project was designed with privacy in mind and uses simulated records for development and analysis. It combines the data-processing and model layers with backend storage and an interactive dashboard so the results can be inspected in one workflow.

What I took from it

The project let me work across the full path from data processing and anomaly detection to API design and visual exploration. It also made the distinction between a useful health-data prototype and a clinically validated tool explicit.