Predictive Analytics for Property Valuation Using Random Forest in Malang City

Abstrak

The property market in Malang City continues to expand alongside rising housing demand, yet limited price transparency still constrains informed decision-making for buyers, sellers, and developers. This study develops a data-driven property price prediction model using the Random Forest algorithm, selected for its robustness and ability to capture complex nonlinear relationships. An initial dataset of 4,358 property listings was collected through web scraping from Rumah123.com, and after thorough preprocessing including data cleaning, handling missing values, and feature refinement 1,573 valid observations remained for analysis. The model incorporates key property characteristics, covering temporal variables (month, year), physical attributes (land area, building area, number of bedrooms and bathrooms, electricity capacity, number of floors), property characteristics (certificate type, property type, property condition, furniture condition, hook position), and price information. Using optimally tuned hyperparameters, the final Random Forest model achieved an R² of 76.66% and a MAPE of 25.27%, indicating strong predictive performance relative to standard regression benchmarks. These findings offer managerial implications by providing objective, data-driven price estimates that can support developers, agents, and prospective buyers in pricing decisions, marketing strategies, and fair value assessments during negotiations.