PRODUCT DEMAND FORECASTING WITH CONSUMER BEHAVIOUR CLUSTERING

Accurate demand forecasting and consumer behaviour analysis are vital for optimizing retail operations, yet
traditional methods struggle with complex, non-linear relationships among pricing, promotions, seasonality, and
external factors. This paper presents an integrated data-driven framework combining demand forecasting with
consumer behaviour clustering.
The system applies ensemble models — Random Forest, Gradient Boosting, XGBoost, and LightGBM — for robust
prediction, alongside K-Means clustering to segment consumers into groups such as price-sensitive, high-value, and
seasonal buyers. Feature engineering and temporal transformations further enhance model performance.
A Streamlit-based analytics platform enables real-time visualization, model evaluation, and user-driven demand
prediction. Results show ensemble methods outperform individual models, while clustering yields actionable
behavioural insights.
By unifying forecasting and segmentation, the framework supports smarter inventory, pricing, and marketing
decisions. It is scalable, efficient, and extensible toward deep learning and real-time data integration.