Estimated Missing Data on Multiple Linear Regression Using Algorithm EM (Expectation-Maximization) for Prediction Revenue Company

Abstract

This study aims to identify problems in the company which in the past two years have experienced unpredictable fluctuations in revenue and do not seem to improve the overall trend in two years. This research is a type of quantitative research involving continuous data types and using 20 independent variables and 1 dependent variable. Data collection was carried out at companies in Sukabumi, Indonesia with data collection methods using interviews and documentation. In the process of data collection, there are some missing data. Based on the type of data that is retrieved, the selection method used in this research is divided into two methods, Expectation-Maximization algorithm for estimating missing data and Multiple Linear Regression to determine the effect of variables on company revenue and to predict the company revenue in the next period. The results showed that using EM Algorithm using the reference variable that has the same character can be predicted the missing data from the initial data. From the calculation of hypothesis test can be concluded that only Luggage, Cleaning Aid, and Glass Ware which have no significant effect to the company's revenue; While the calculation using multiple linear regression analysis there are 5 types of goods that have a negative influence with company revenue, the 5 types are: Hair care, Luggage, Bakery, Cleaning Aid, and Snack. Thus, the overall type of goods that are very necessary to be evaluated by the company are: Luggage, Bakery, Cleaning Aid, Snack and Glass Ware, because of the type of goods that do not have significant influence or negatively effect to the increase of the company's revenue.