Fusing Clustering and Machine Learning Techniques for Big-Mart Sales Predication

Abstract

Sales volume forecasting is becoming a crucial requirement for many sellers, enabling them to modify their advertising strategies and strike the ideal balance between sales volume and profit. A higher budget for promotion and a lower profit are the outcomes of overestimating the sales volume, which results in lower actual sales than predicted. Therefore, accurate sales volume forecasting is essential for sellers to choose the best methods of advertisement. The quantity of sales predicting approach in this case is one that predicts future sales based on historical data for the particular supplier, as well as the advertising techniques to be used on the relevant future days, sales volume from the past, dynamic influencing elements like past marketing strategies and weather data, and the static characteristics of the specific vendor are all examples of historical information. As a result, merchants use the expected forthcoming gross revenue under various stuff to choose the deal strategy to be employed. This work suggests a clustering-based forecasting methodology for predicting sales by fusing clustering and machine-learning techniques. The suggested method first divided the training data into groups using the clustering technique, which groups data with comparable features or patterns. The forecasting models for each group are then, using machine learning techniques, it was trained. Once, the cluster's trained forecasting model was chosen for sales forecasting. It was determined which cluster's data patterns matched the test data the most closely. Because sales data reflect similar data patterns or features over time, the presented clustering-based forecasting methodology can enhance forecast accuracy. Three machine learning techniques are used in this paper: Generalized Linear Regression, Decision Tree, and Gradient Boosted Tree, as well as two clustering techniques: Self Organizing Map (SOM) and K-mean clustering. A predictive model that is most suited for predicting the sales trend is presented based on a performance evaluation. Discussion of the data is done in terms of the precision and accuracy of good prediction systems. The best-fit model is chosen to be the Gradient Boosted Tree Algorithm and it has the highest level of forecasting and sales prediction accuracy. The sales forecast, or which product will be sold in the future and in what quantity, is carried out using Gradient Boosted Tree.