Sales Prediction Based on Lasso Regression

Abstract

Sales prediction is a critical aspect for businesses across diverse fields, providing them with the means to operate efficiently and achieve success. It constitutes an integral component of the decision-making and planning processes within a business. Several forecasting models are available for sales prediction, with most machine learning models performing exceptionally well. However, the suitability of these models can vary depending on the dataset provided. While many datasets are not overly complex and contain a limited number of variables, others are more intricate, featuring numerous variables, many of which may be irrelevant and could potentially skew the results. Therefore, the goal is to eliminate these irrelevant variables and identify those that are more closely correlated with the prediction task. The Least Absolute Shrinkage and Selection Operator (LASSO) regression model emerges as a valuable tool for the removal of irrelevant data. This study employed the Lasso regression model to analyze a housing sales dataset and discovered that it provided an excellent fit for prediction without overfitting the training set. This study presents an overview of the advantages of LASSO, discusses its limitations, and offers insights into its potential for future development.