Machine Learning in the Prediction of Net Sales for Colombian Companies in a Post-pandemic Scenario

Abstract

 Abstract — The uncertainty about how the economic reactivation will behave worldwide is a general concern; in the face of this panorama, it is essential to look for historical data that allow us to build the present and predict the future, with this purpose and taking advantage of the advancement of technology in the field of Machine Learning, the present work established the predictions on net sales by companies operating in Colombia. To this research, about two million official records were used from the open data portal of the Bogotá Chamber of Commerce, which were divided 70% for training and 30% for tests; based on these data, Linear Regression algorithms were used (LR), Random Forest (RF), XGBoost (XGB), and Extreme Learning Machine (ELM) to make predictions. The results of the regression performance were evaluated through the coefficient of determination, and the best measure performance showed 0,9 with a Random Forest regressor (RF)