Riot Games Project

End-to-end machine learning platform for League of Legends analytics with a FastAPI backend, React frontend, Riot API integration, and MLflow experiment tracking.

Group Members

• Khalil Ayeb
• Melek Bouabid

Objective

Build and evaluate machine-learning models using Riot API match data to predict:

• match outcomes
• player rank tier
• smurf anomalies
• rank progression

while identifying which gameplay features most influence these predictions.

Data Schema

See data/processed/README.md for the updated feature dataset schema.

Progress Summary (Feb 27, 2026)

✅ What We Achieved

• Trained and integrated 4 core prediction modules:
  • Rank Tier Classification
  • Progression Regression
  • Smurf Anomaly Detection
  • Match Outcome Prediction (multiple variants)
• Built a production-style API layer with FastAPI (/api/v1).
• Connected frontend pages to live backend endpoints.
• Added Riot API resilience features (cache + retry/backoff + health status mapping).
• Added UI/UX improvements: skeleton loading, retry cooldown, status badges, cache age and timestamps.

MLflow

Current Setup

• MLflow is used to track experiments, metrics, artifacts, and model runs.
• Local tracking data is stored under mlruns/ and artifacts under mlartifacts/.
• Setup and comparison scripts are available in scripts/3_model_training/.

Run MLflow UI

mlflow ui --port 5000

Open: http://127.0.0.1:5000

MLflow Scripts

• scripts/3_model_training/mlflow_setup.py → initialize experiments
• scripts/3_model_training/train_all_models.py → train models with tracking
• scripts/3_model_training/mlflow_model_comparison.py → compare runs

Model Choice

Selected Model Families

• Rank Tier Classifier
  • Predicts rank tier from player performance features.
• Progression Regressor
  • Predicts win-rate/skill progression tendency.
• Smurf Detector
  • Anomaly-based detector for suspiciously high-skill behavior.
• Match Outcome Predictor
  • Supports early, full, strict, and cascade variants.

Why This Choice

• Covers classification + regression + anomaly detection for complete player analysis.
• Enables both offline evaluation and live API inference.
• Match outcome variants support different feature availability windows.

FastAPI Setup

Environment

• Python dependencies: requirements.txt
• Settings source: environment variables in .env
• Main app entry: api/main.py
• Config object: api/core/settings.py

Start Backend

.venv\Scripts\python.exe -m uvicorn api.main:app --host 127.0.0.1 --port 8001

API Base

• Base URL: http://127.0.0.1:8001
• Prefix: /api/v1
• Docs: /docs
• OpenAPI JSON: /openapi.json

Endpoint Implementation

Health

• GET /api/v1/health
• GET /api/v1/health/riot

Match Outcome

• GET /api/v1/match-outcome/models
• POST /api/v1/match-outcome/predict/early
• POST /api/v1/match-outcome/predict/full
• POST /api/v1/match-outcome/predict/cascade
• POST /api/v1/match-outcome/predict/strict

Other ML Endpoints

• POST /api/v1/rank/predict
• POST /api/v1/progression/predict
• POST /api/v1/smurf/predict
• POST /api/v1/summoner/predict

Riot Resilience Implemented

• Response cache for summoner prediction endpoint.
• Retry with exponential backoff for transient Riot failures.
• Health mapping for key states (active / expired / rate-limited / unreachable).

Swagger Testing

How to Test in Swagger

1. Start backend server.
2. Open http://127.0.0.1:8001/docs.
3. Expand an endpoint (for example POST /api/v1/rank/predict).
4. Click Try it out.
5. Paste a valid JSON payload.
6. Click Execute and inspect response + status code.

Recommended Quick Checks

• GET /api/v1/health
• GET /api/v1/match-outcome/models
• POST /api/v1/summoner/predict (requires valid Riot API key)

Frontend

Stack

• React + TypeScript + Vite
• Routing with React Router
• Charts/visualization with Recharts

Start Frontend

cd "frontend/hextech-insights (1)"
npm install
npm run dev

Open: http://127.0.0.1:5173

Recent Frontend Improvements

• Removed duplicate summoner search UI in Live Analytics flow.
• Added model-card navigation from Model Dashboard to Predictions page.
• Added a compare-players page with side-by-side metrics.
• Added “Why This Prediction” explainability cards.
• Added loading skeletons, retry cooldown, empty states, Riot status badge, and update timestamps.

Frontend-Backend Integration

Integration Status

• Frontend API client is centralized in frontend/hextech-insights (1)/services/api.ts.
• Uses VITE_API_BASE_URL (default: http://127.0.0.1:8001/api/v1).
• Predictions and profile flows are linked to backend endpoints and typed responses.

Data Flow

1. User action from frontend page.
2. Frontend calls FastAPI endpoint.
3. Backend loads model/service and returns prediction payload.
4. Frontend renders metrics, charts, explanations, and status indicators.

Next Priority

• Replace expired Riot dev key with active key to fully validate live summoner pipeline end-to-end without fallback mode.
• Add Docker Compose for one-command startup (backend + frontend + MLflow).