ML Development & Engineering for Hong Kong Horse Races

Objective: To develop a predictive model for race day outcomes that lays the groundwork for a real-time betting system.

Developed efficient data processing scripts and enhanced feature engineering with TsFresh, adding over 25 critical features to the dataset.

Conducted classification analysis with accuracies ranging from 65% to 90% for different outcomes, optimizing bet amounts with Hyperopt.

Orchestrated an AWS-based pipeline with MySQL for data storage, delivering real-time predictions every 30 seconds via Flask.

Challenges & Outcome: Although the models achieved high accuracy levels, they exhibited occasional prediction variability, largely due to data-related challenges like extensive missing values and scarcity of data. This variability impacted the revenue generation to some extent.