Comparative Analysis of Machine Learning Algorithms for Sales Forecasting in the Russian Toy Retail Sector

This research provides a detailed examination of various machine learning approaches for the sake of forecasting sales within the Russian toy retail industry. This sector is marked by fluctuating consumer preferences and a pressing need for precise inventory control, making accurate sales predictions vital. The study aims to determine which forecasting technique yields the most reliable results for this dynamic market. To achieve this, several predictive models were assessed, including Linear Regression, Random Forest, Light Gradient Boosting Machine (LightGBM), and Extreme Gradient Boosting (XGBoost). The analysis involved a comprehensive process encompassing data preprocessing, exploratory data analysis (EDA), feature engineering, and rigorous model training, using a dataset sourced from Kaggle. The results highlight that Random Forest consistently delivers superior predictive performance compared to its counterparts. This model excels in balancing accuracy with generalization, making it particularly effective for sales forecasting in environments characterized by variability and uncertainty. The practical implications of these results are substantial, offering retailers a robust tool for refining their operational strategies, optimizing inventory management, and enhancing customer experience through improved forecasting accuracy.