運用正規化迴歸分析線上銷售預測： 以在亞馬遜上的運動商品為例

Online Sales Forecasting by Regularized Regression for functional products: Taking Sport Goods on Amazon.com as an Example.

劉育吟(Yu-Yin Liu)

指導教授 ： 孔令傑

國立臺灣大學/管理學院/企業管理碩士專班/碩士(2020年)

https://doi.org/10.6342/NTU202000275

摘要

無資料

關鍵字

電子商務 ； 銷售預測 ； 機器學習 ； 套索回歸 ； 嶺回歸

並列摘要

Abstract
In this study, we research on a company’s sport good sales forecasting on Amazon.com. We analyze data including transactions, advertisement reports, customer reviews, competitors’ prices and customer reviews, holiday-or-not, and weekend-or-not for more than 500 days. We implement machine learning models to tackle the sales forecasting problem. The main objective of this study is to discover the most efficient model among linear, LASSO, and Ridge regression by comparing their mean absolute error in the testing set. We find that the most efficient model is LASSO regression in general, whose performance may be better than linear regression by 87 % on a certain product.

並列關鍵字

e-commerce ； sales forecasting ； machine learning ； LASSO regression ； Ridge regression

參考文獻

Bibliography

Ahn, J.J., H.W. Byun, K.J. Oh, and T.Y. Kim. (2012). Using ridge regression with genetic

algorithm to enhance real estateappraisal forecasting. Expert Systems with

Applications, pp. 8369-8379.

Amazon. (2019). FBA inventory storage limit. Retrieved from Amazon Seller Central: