Optimizing E-Sports Revenue: A Novel Data Driven Approach to Predicting Merchandise Sales Through Data Analytics and Machine Learning

This research work presents a comprehensive approach to predicting merchandise sales in the rapidly growing E-sports industry. Utilizing a rich dataset for data analytics, comprising of over 450 games’ historical records, including the number of players, earnings, tournaments count, and merchandise profit, robust Machine Learning models were developed to forecast future merchandise sales. We employed several regression techniques, including the Huber Regressor, CatBoost, and HistGradientBoostingRegressor, each providing unique benefits. The Huber Regressor excelled in handling outliers, ensuring the model did not overfit. CatBoost handled categorical variables effectively, while HistGradientBoosting Regressor demonstrated high performance with large datasets. Furthermore, the models were tuned using a combination of hyperparameter optimization and cross-validation methods. Our Huber Regressor model, after tuning with an optimal epsilon of 1.0, achieved a nearly perfect R \(^2\) score. However, due to the risk of overfitting, cross-validation was used, yielding a Root Mean Squared Error (RMSE) of approximately 0.0025, indicating high precision. The CatBoostRegressor model, while not as precise as the Huber Regressor, was notable for its flexibility in handling categorical variables. After tuning and cross-validation, the model yielded an R \(^2\) score of approximately 0.42. TheHistGradientBoosting Regressor model also performed well with an R \(^2\) score of 0.776 on the test set, indicating a good fit to the data. These models, through careful data pre-processing and tuning, demonstrated the potential to accurately predict merchandise sales, enabling stakeholders to optimize revenue generation. We also conducted scientific analyses using LIME, SHAP, and ELI5 to gain in-depth insights into the data. This work underscores the value of utilizing machine learning and statistical testing for decision-making, offering insights and recommendations for driving revenue growth in the E-sports sector. Further research could explore additional features and models, building on the foundation laid in this project.