Sales Prediction Scheme Using RFM based Clustering and Regressor Model for Ecommerce Company

Machine learning models are being used for better insights and decision making across many industries today. It shows to be quite useful for businesses in the ecommerce industry as well due to the vast amount of data generated and its potential. This research aimed to find insights on future sales of an ecommerce company [1]. The vast number of variables including both categorical and continuous variables under product data, customer information, transaction information, led us to implement a prediction model using regressors rather than just time series forecasting techniques. First an RFM (Recency, Frequency and Monetary) based clustering algorithm was used to get customer related information and then integrate those results into a regressor to achieve the desired goal of prediction of sales. Two schemes were tested one being predictions on individual clusters and the other where the clusters were one hot encoded back into the main data. Results show quite high accuracy of prediction. The high R-squared also indicated that our hypothesis of including the variables contributed significantly to the predicted sales values was correct in this case. This research fulfills an identified need to understand how machine learning algorithms can be implemented by multiple algorithms being integrated in sequential and logical orders thus helping derive business specific strategies rather than making it a mere technical process by providing empirical results about how the predicted sales values along with given inputs can contribute in business decision making relating to marketing, inventory management, dynamic pricing or many more such strategies.