Members:
- M. Callao Pimentel.
- E. P. Gonzales Requejo.
- C. Gutierrez Bravo.
- G. Seminario Paredes.

Abstrac: Estimating the price per square meter of real estate properties in Lima Metropolitana presents significant
challenges due to the city’s dynamic and heterogeneous market conditions. Both buyers and sellers often face
difficulties establishing an accurate market price, leading to inefficient transactions and dependence on costly
intermediaries. This research addresses this issue by applying machine learning techniques to predict the price per
square meter of housing properties, using historical data from public databases provided by the Central Reserve Bank
of Peru (BCRP). Initially, extensive exploratory data analysis (EDA) was performed to identify and correct outliers,
impute missing values, and assess variable correlations and relevance. Seven regression models were evaluated:
Linear Regression, Polynomial Regression, Decision Tree, Random Forest, SVR, XGBoost, and Multilayer
Perceptron (MLP). Variables were encoded using One-Hot Encoding, and MinMax scaling was applied specifically
for SVR and MLP models. Performance was assessed through MAE, MAPE, RMSE, and R² metrics. The Random
Forest model without exogenous variables demonstrated superior predictive performance, achieving a remarkably
low MAE (8.47) and an exceptionally high R² (0.9996). This study highlights the effectiveness of Random Forest for
property valuation tasks, providing a robust and practical tool for enhancing transparency and reducing transaction
costs in Lima’s real estate market.

Keywords: rice prediction, real estate, regression models, Random Forest, Lima Metropolitana, machine learning.