Konsistensi Model Regresi Empat Variabel Pada Populasi dan Sampel untuk Prediksi Temperatur

Abstract

The ability to predict future events or trends has become very important today. One method that can be used to predict the future is to use linear regression. Accurate regression modeling requires sampling representative data, especially when working with large datasets. This research takes a relatively large volume as a data set by looking at the accuracy and consistency of the coefficients of a multi-variable linear regression model for temperature prediction which is built based on all the data, and looks at the differences in the regression model built from the sample data. The number of sample data (n) is determined based on the Slovin formula which depends on the number of population data (N) and the level of confidence (ơ), so that as many as (N/n) new regression models can be built. Each group of sample data is divided into 75% for modeling and 25% testing data. The dataset used is weather information in the Szeged area which was measured in 2006 - 2016. So the regression model is in the form of Y (temperature value) which is influenced by Xi (weather factors), namely humidity, wind speed, wind direction and visibility. Using 96,453 data records and a 1% significance level in Slovin's formula, 10 samples were generated. Nine out of ten sample regression models agree with the population model, with positive coefficients for visibility and wind direction and negative values for humidity and wind speed. There is an abnormality in sample #4. While the other nine sample regression models are consistent with positive R²  values, Sample #1 displays an oddity with negative values. The RMSE interval values for each regression model in this study fall between 4.334 and 9.582.