Prediction of Future Sales Using Machine Learning Algorithms

In this work, forecasting of future monthly sales have been carried out. Available resources can be properly utilized with the help of accurate forecast. For this work, AutoRegression Integrated Moving Average (ARIMA), seasonal ARIMA (SARIMA) and extreme gradient boosting (XGBoost) machine learning algorithms have been used to build the forecasting model of monthly sales of products of different stores. Root mean square error (RMSE) has been used as performance metrics to decide the best performing model. ARIMA is the worst performing forecasting model for every product because monthly sales variations have some seasonality component in it. To decide best ARIMA or SARIMA model minimum AIC value has been used as criterion. XGBoost is used to build forecasting model using its regression algorithm using day, month and year as input features and monthly sales as target variable. The performance of SARIMA forecasting model is better than XGBoost model for some products and vice-versa. XGBoost is concluded as the best forecasting model for this study because it can handle multivariate data and other features can also be incorporated in future.