Development of a predictive statistical system for sales production: a case study of retail shops in Wakiso, Uganda.

Abstract

In today's competitive market environment, accurately predicting sales is essential for effective inventory management, marketing strategies, and overall business planning. This project aims to develop a robust sales prediction model by leveraging a variety of influential factors including seasonal trends, promotional activities, holidays, and weather conditions. Utilizing the Random Forest algorithm, known for its ability to handle complex and non-linear relationships, the model is trained on historical sales data enriched with these additional predictors. The dataset comprises daily sales records, with supplementary information on whether promotions were active, if the day was a holiday, and the prevailing weather conditions. Each sales record is further processed to extract the month and day of the week, capturing temporal patterns that affect consumer behavior. These factors collectively enable the model to learn from historical trends and improve the accuracy of sales forecasts. A Shiny web application is developed to provide an interactive platform for users to predict sales for any given date by inputting relevant factors such as promotions, holidays, and weather conditions. The application not only displays the predicted sales but also visualizes sales trends over time, aiding users in understanding the influence of different predictors. Initial results indicate that incorporating these additional predictors significantly enhances the model's performance, offering more precise sales forecasts compared to models relying solely on temporal data. This project demonstrates the potential of integrating diverse data sources to build more accurate predictive models, ultimately contributing to more informed business decision-making.