The goal of this project is to predict which horse in a race can be the first 3 winners.

The data is given and can be downloaded here

One should have at least basic understanding of horse racing in order to extract information from the given dataset.

Basic information about Japanese horse racing can be found here

All dependencies are listed in the requirements.txt

download requirements.txt

In the terminal, type: pip install -r requirements.txt

• notebook_data_processing: preprocess and select meaningful data to build model
• classification: horse top3 prediction based on preprocessed data
• Input file: data is given in the compressed file given in the Data section. After unziping it, data in historical_data folder will be used.

• Download the data in the link given above.
• extract and place file in folder data/ in the main directory

• In ./src/data_process, run data_process.py file to process raw data.
• The outputs are training.csv and testing.csv in extract_feature folder

• In the newly created training and testing files, there are still many features for modeling. Therefore, several methods are applied to choose suitable features.
• Run the file feature_selection.py then it will output useful_features.json in the same directory.

• Logistic Regression and Random Forest models are built and tuned in the basic_prediction_model.py.
• CatBoost and LightGBM models are built and tuned in advanced_pred_model.py. Moreover, soft and hard ensemble models are also created based on CatBoost and LightGBM.
• Scores of all models are recorded in src/models/score dicrectory.
• Optimal parameters of each model are place in separated json files in src/models/best_hyperparameters directory.
• After having tuned all parameters, training and testing datasets are combined and used to build soft voting ensemble in final_model.py. It output *_final.pkl models.

• Working on it

https://github.com/GarlicSoup/horse_racing_prediction

This program is created by Hieu Le