Perbandingan Analisis Faktor Penentu Penjualan PT. X Menggunakan LASSO Regression dan Gradient Boosted Regression Tree

Abstract

Information becomes a crucial asset for an organization. However, employees of PT. X are facing difficulty in analyzing data because it has to be processed one by one. Moreover, analyzing data in an operational database is not recommended as it can interfere with the performance of the operational database. Then, when the Board of Directors want to know the reason behind its sales’ performance, they conclude it based on their mere assumption. This research implemented a data warehouse with the help of ETL tools. Then, sales transactions of PT. X were analyzed to get information about factors that affect company’s revenue. Factor models were formed for brands which sales were not good enough these past few years. Factors which are examined are sales price, stock availability, on time delivery of goods, quantity of returns, month of transaction, and cost price. The analysis was carried with two methods, LASSO regression and Gradient Boosted Regression Tree. These models were measured by Root Mean Squared Error, R-squared, and Variance Inflation Factor to know which model performs better. Result of the research shows LASSO regression and Gradient Boosted Regression Tree succeed in performing feature selection for sales transactions of PT. X. Yet, the factor model from Gradient Boosted Regression Tree gives a better result than LASSO regression. Last, a program was made for the company in the need of future analysis using Gradient Boosted Regression Tree.