ROBUST REGRESSION MODEL AND ITS SIMULATION ON PROBLEM OF OUTLIER

From previous research, population inference could be improved using non-sample prior information (NSPI). The NSPI is often used to test the parameter on linier model (regression model) in estimation of the population inference. In the case of outliers, the technique on population inferences could be approached using robust regression model. In this research, we focused on comparing the performance of robust regression methods, specifically the least trimmed squares (LTS) and Scale (S) using the Tukey bisquare weighting function, and M-estimation using both Huber and Tukey bisquare weighting functions. Here, we used two kinds of the data, namely generated data from R software and the real data about inflation data from Indonesia 1990-2024 (including the 1997-1998 crisis period) with economic growth, interest rates, and foreign exchange rates as predictor variables. For the first data, the result showed that the model with the S has an adjusted R-squared value of 86.6% and an MSE value of 450.23. While the model with the LTS has an adjusted R-squared value of 80.4% and an MSE value of 453.33. Thus, we recommended that the S method is an eligible choice. In the case of real data (second data), the model in term of the Huber function has an adjusted R-squared value of 91.2% and an MSE value of 8.49. While the model with the Tukey bisquare function has an adjusted R-squared value of 88.8% and an MSE value of 11.88. In conclusion, robust regression using the S and Huber is highly effective in handling outliers and improving the reliability of population inference.