MatchPulse Live

• 

  List other stats

• 

  List raw

• 

  List stats

• 

  Match screen

• 

  Logo

• 

  Coefficients

Inspiration

Live football apps show scores, but they don’t explain what is happening or what might happen next. We wanted to turn raw match data into actionable insights.

What it does

MatchPulse Live analyzes real time match data and provides:

• Win/draw probabilities using a machine learning model
• A smart formula-based prediction for comparison
• Goal threat detection to highlight matches likely to have a goal soon
• Identification of "hot games" with high intensity

How we built it

We built a Next.js application that consumes live football data via API.

We used Zerve to:

• Generate synthetic match-state datasets
• Train a multinomial logistic regression model
• Extract feature importance and coefficients

The trained model is then integrated into the frontend to provide real time predictions.

Challenges we ran into

• Working with incomplete and inconsistent live API data
• Designing meaningful features from raw match stats
• Handling API request limits efficiently
• Training a usable model without access to historical labeled datasets

Accomplishments that we're proud of

• Real time prediction system running on live matches
• Integration of ML and rule-based models side-by-side
• Clean UI highlighting key insights instantly
• Goal threat detection logic

What we learned

• Feature engineering is more important than model complexity
• Real time systems require careful API usage and caching
• Synthetic data can be useful for prototyping ML systems

What's next

• Train the model on real historical match data
• Improve draw prediction accuracy
• Add momentum tracking over time
• Deploy the system as a public API

Built With

• football-api-(live-data)
• next.js
• python
• tailwind-css
• typescript
• zerve-ai-(ml-training)