Abstract:

The AmazonSalesPredictor model is a reliable predictive model designed on efficient statistical methods to boost the accuracy rates of Amazon's large volume data set. This paper proposed a Hybrid Model including LSTM networks and GBDT to provide accurate results when capturing temporal dependencies and feature interactions. LSTM captures the relationship between the data points in sales data as it is sequential while the GBDT compels the model to identify the relation between the features for precise prediction. The proposed model yielded an MSE of 228202.57, a MAPE of 0.04, and a score of 0.97 hence maintaining a high fit accuracy. The former of these indicators shows how the model achieves very accurate sales predictions with little margin for error and is in line with the observed data on sales. Further, an easily navigable Power BI dashboard also augments the model through an ability for the stakeholders to analyze predictions by year, by quarter, and by-product genre. This format enables the users to interact with the displayed data in several ways to get a detailed perception of sales performance, profitability, and product competitiveness for various sub-categories and geographical locations. The AmazonSalesPredictor is a useful and valuable BI tool as it allows for strategic medium and long-term business decisions on stock amounts, sales promotion, and budgeting. As a result, owing to that amalgam approach and richer visualization abilities, the model meets sophisticated demands in the sphere of predictive analytics, allowing Amazon to operate proactively to the dynamics of markets and customers’ demands’ shifts.