AI-Powered Cigarette Demand Forecasting: A Hybrid Machine Learning and Deep Learning Approach

This study puts forward a framework driven by AI, aimed at forecasting cigarette sales through combining various machine learning and deep learning models in a bid to enhance the precision and interpretability of forecasts. Different algorithms are evaluated, which includes models like Random Forest, Gradient Boosting Trees, XGBoost, LightGBM, Support Vector Machine, and Artificial Neural Networks, to make predictions about cigarette demand. It is indicated that models such as RF, GBT, XGBoost, and LightGBM offer forecasts that are quite accurate overall, displaying low RMSE and high $\mathbf{R}^{2}$ values, hence they become the selected models for this forecasting endeavor. Analysis surrounding feature importance indicates that economic factors, including aspects like product category, income, and GDP, are significant to consider when predicting cigarette sales. The benefits of employing an ensemble-based approach are emphasized, which combines different model strengths in pursuit of greater robustness and adaptability, particularly in rapidly changing market situations. This framework holds insights that might be of value to stakeholders involved in the tobacco industry, potentially assisting in the refinement of marketing and distribution strategies that they use. There could be future research that thinks about integrating real-time data along with enhancements in deep learning methods for better accommodating the complexities of the market.