Implementation of Machine Learning using the Multiple Linear Regression Method for Boarding House (Kost) Rental Price Prediction Around Universitas BSI Jatiwaringin Campus

Abstract

SIPREKOS is a web-based boarding house (kost) rental price prediction system developed to help students and kost owners around Universitas BSI Jatiwaringin Campus obtain objective price estimates. High rental prices and limited price transparency motivate the need for such a system. This study builds a prediction model using the Multiple Linear Regression method with polynomial features, based on nine kost predictor variables together with a squared-distance derivative variable to capture the non-linear effect of distance on price. Primary data were collected manually from 115 kost units listed on Mamikos.com within a 3 km radius of the campus during May–June 2026, then cleaned using an absolute-error-based residual analysis method, resulting in 100 valid records. The valid data were split 80:20 into training and test sets; regression coefficients were estimated from the training set using the Normal Equation. Testing shows the model explains 72.4% of rental-price variance (R² = 0.724) with an average prediction error of IDR 116,969 (MAE), and it passes the classical assumption tests (multicollinearity, heteroscedasticity, and normality), confirming that the estimator is BLUE (Best Linear Unbiased Estimator). Black-box testing across ten scenarios shows that all system functions, for both general users and administrators, operate according to specification.