藉由LASSO演算法選擇迴歸變數

Abstract

Abstract

This study considers using the LASSO (least absolute shrinkage and selection operator) method to select important independent variables in linear regression models. The LASSO method is a supervised learning algorithm. By constraining the properties of the residuals, an interpretable regression model is developed, and the constrained parameters are set to be zero. This report discusses how to appropriately select important variables via a simulation study when there are too many independent variables or even the number of independent variables greater than the number of sample size. We use simulated data in 500 replications to show how the LASSO method selects important variables of the regression model. Finally, the results of simulation data are provided, which show that almost all important variables and half of the correct combination of variables can be accurately selected.

摘要

在許多數據集中，包含的變量動輒上百甚至更多，這使得我們必須適當選取變量以及降維技術，使得我們可以在最大程度發揮模型的解釋能力。本研究考慮使用LASSO（最小絕對收縮和選擇算子）方法在線性回歸模型中選擇重要的迴歸變數。 LASSO模型是一種監督學習算法。通過約束殘差的屬性，提出可解釋的迴歸模型，並將約束參數設置為零。本報告討論了當自變量過多，甚至迴歸變量的數量大於樣本數量時，如何通過模擬研究適當選擇重要變量。我們重複500次中的模擬數據來顯示LASSO模型如何選擇迴歸模型的重要變量。最後，提供了模擬數據的結果，表明幾乎可以準確選擇所有重要變量和變量正確組合的一半。