Optimize Existing ML Code

I have an existing ML code that currently works as follows:
1.) Gets football data (including updated data) from https://www.football-data.co.uk/
2.) Clean and preprocess the data
3.) Create additional features and targets
4.) Loops through and train/test/validate each target
5.) Obtained a minimum accuracy of 90%
6.) Predicts upcoming matches using the trained models

ISSUES:
1.) Despite the alleged 90% accuracy (on seen data), the model performs poorly on unseen data
2.) It appears to have leaked features
3.) I suspect overfitting

TASK:
1.) Check for either of the issues listed above
2.) Fix the identified issues
3.) Optimise the code for better accuracy and performance.

Timeline: Urgent
Budget: Approx. $20