AI-Driven Risk Profiling & Anomaly Detection

An intelligent health reporting system designed for HackArizona 2026. This platform leverages professional health reporting and historical CDC datasets to identify emerging health anomalies and risk profiles in real-time.

🚀 Overview

This project provides a bridge between individual user health reporting and macro-level CDC data. By cross-referencing recent professional diagnoses with historical trends, the system identifies statistical anomalies that could indicate localized outbreaks or health trends.

• User Reporting: Securely captures professionally diagnosed illnesses from the past month.
• Anomaly Detection: Uses AI services to compare current reporting rates against baseline CDC data.
• Risk Profiling: Generates dynamic risk scores and visual trends for specific regions/ZIP codes.

---

🛠️ Tech Stack

Frontend

• Framework: React 18 with TypeScript
• Build Tool: Vite
• Styling: Tailwind CSS + shadcn/ui
• State Management: Custom hooks & local state logic
• Testing: Vitest

Backend

• Language: Python 3.11
• Framework: Flask
• Database: SQLite (SQLAlchemy)
• AI/Analysis: Custom Anomaly & AI Explanation Services
• Containerization: Docker & Docker Compose

---

📊 Information Sciences Approach To Anomaly Detection

The anomaly engine uses a four-step composite scoring model that combines local report trends with CDC FluView baseline data to generate a risk score between 0 and 100.

Step 1 — Percent Change

$$pct_change = ((this_week - last_week) / last_week) × 100$$

Compares report volume this week against last week. If last week had zero reports, the system returns a maximum signal of 999%.

Step 2 — Z-Score

$$z = (this_week - last_week) / (cdc_std × 3 + 1)$$

Measures how statistically unusual the current spike is, anchored to the CDC FluView standard deviation for the matching epiweek.

Step 3 — Trend Signal

Condition	Classification	Signal
Last week = 0, this week > 0	Accelerating	75.0
Ratio > 1.5	Accelerating	(ratio − 1) × 100
Ratio > 1.05	Increasing	(ratio − 1) × 100
Ratio ≥ 0.95	Stable	0
Ratio < 0.95	Declining	0

Step 4 — Composite Score (0–100)

$$score = (0.25 × pct_signal) + (0.35 × z_signal) + (0.20 × trend_signal) + (0.20 × cluster_signal)$$

Where:

• pct_signal = min(100, pct_change / 3)
• z_signal = min(100, |z| × 25)
• cluster_signal = min(100, this_week × 5)

Z-score carries the highest weight (35%) as it is the most statistically grounded signal. Percent change follows at 25%, with trend direction and cluster size each contributing 20%.

Step 5 — Risk Classification

Score	Label
0 – 25	🟢 Normal
26 – 50	🟡 Emerging Anomaly
51 – 75	🟠 Significant Anomaly
76 – 100	🔴 Possible Outbreak

📂 Project Structure

HackArizona2026/
├── backend/             # Flask API, AI Services, & SQLite DB
│   ├── models/          # Database schemas (SQLAlchemy)
│   ├── routes/          # API Endpoints (Reports, Anomalies, Health)
│   ├── services/        # Core logic: AI Explanation & Anomaly Detection
│   └── data/            # Seed data and historical CDC records
├── frontend/            # React + Vite application
│   ├── src/components/  # High-level UI (RiskScoreBar, TrendBadge, etc.)
│   ├── src/pages/       # Dashboard and ZipDetail views
│   └── src/lib/         # API wrappers and utility functions
└── docker-compose.yml   # Orchestration for full-stack deployment