Additive Tuning Lasso (AT-Lasso): A Proposed Smoothing Regularization technique for Shopping Sale Price Prediction

Abstract

In this paper, we developed a prediction model for Shopping Sales Data especially Black Friday sales. This model is used during Black Friday day because that day sales hugely vary from normal day sales. Black Friday deal dependent on various variables includes Age, Marital Status, Occupation, Product categories, Duration of Stay in the Current City, Gender, and City Category. The number of methods was implemented which include Linear Regression, Lasso Regression, Elastic Net Regression, and Ridge Regression for predicting sales. The choice of Regularized methods to be considered to perform a prediction model in this study. However, these methods fail to produce optimal features that are active. Also, these methods limit to model with linear features. The proposed method focused on these issues and resolved by extending general regularized Lasso with Tuning Parameter and Additive Models called Additive Tuning Lasso (AT-Lasso). A model that focused on identifying active set with both linear and non-linear features. The performance of method compared against standard regularized methods Lasso, Ridge, and Tuning-Lasso with benchmarks of MSE, DF and computation time. The results shown proposed is promising among standard methods.