Prediksi Harga Tanah menggunakan Machine Learning Algoritma Regresi Linier

Abstract

Conventional land price determination in Surakarta City remains manual, subjective, relying on personal assessments that tend to follow land prices in the surrounding area, and is inefficient for large-scale data. This study aims to implement the Multiple Linear Regression algorithm to predict land prices based on physical and environmental characteristics. The data used comprises 1,923 offer and transaction records for land from 2021 to 2024 sourced from the Surakarta City Land Office. The methodology employed is CRISP-DM, which consists of six stages: business understanding, data understanding, data preparation (including cleaning, normalization, and data splitting into 80% training data and 20% testing data), modeling, evaluation, and deployment. The features used include land area, road width, distance to the city center (derived from X,Y coordinates), building area, number of floors, and road class (one-hot encoding). The results show that the model achieved an R² = 0.7767, RMSE = IDR 684.3 million, and MAE = IDR 503.9 million. The RMSE is smaller than the standard deviation of land prices, IDR 1,493,000,000, indicating that the model is considered good. The model coefficients indicate that building area and land area have the largest positive influence, while distance to the city center has a negative effect. A Streamlit-based prediction prototype has been deployed to facilitate users in inputting land parameters and obtaining real-time price estimates. This study contributes an objective, fast, and easily interpretable prediction model for use by government agencies and property practitioners in decision-making.