League of Legends Win Predictor

Overview

An end-to-end classical ML project: an offline training pipeline produces a StandardScaler + LogisticRegression model from match statistics, and a Flask service serves predictions through a JSON API and an interactive dashboard that also displays the model's validation accuracy and ROC AUC.

Challenge

Match outcomes depend on many in-game statistics. The goal was a simple, interpretable model with a reproducible training pipeline, packaged as a service that others can query and understand.

Solution

A training script that splits the data, fits a scikit-learn pipeline (StandardScaler + LogisticRegression), and saves the artifact with metadata (feature order, target, validation metrics); a Flask app factory with /predict and /health endpoints and a dashboard UI.

Architecture

pipeline/train_pipeline.py → artifacts (joblib + metadata.json) → Flask API (/predict, /health) → dashboard

Results

• Model: Logistic Regression
• API: /predict · /health
• Reproducible, parameterized training pipeline
• Versioned artifacts with feature metadata and validation metrics
• JSON prediction API plus health probe
• Dashboard shows accuracy / ROC AUC alongside predictions

Technologies

• Python
• scikit-learn
• Logistic Regression
• Flask
• Pandas
• joblib