A Machine Learning Framework for Land Price Estimation in Huye District

Accurate land valuation is essential for effective land governance, taxation, urban planning, and in-vestment decision-making, particularly in rapidly urbanizing developing countries such as Rwanda. However, existing land valuation practices in Rwanda remain largely manual, subjective, and inconsistent, limiting efficiency, scalability, and transparency. This study presents a machine learning-based framework for land price estimation in Huye District.

A multi-source dataset integrating spatial and administrative land attributes was constructed from land registry databases, official valuation schedules, Geographic Information Systems (GIS), and administrative land records. The initial dataset contained 51,955 records, which were reduced to 35,511 records after preprocessing and outlier removal. Land price estimation was formulated as a supervised regression problem, and five machine learning models, including Random Forest, Decision Tree, Linear Regression, K-Nearest Neighbors (KNN), and Artificial Neural Network / Multilayer Perceptron (ANN/MLP), were implemented and evaluated.

Experimental results demonstrated substantial differences in predictive performance among the evaluated models. Random Forest achieved the best overall performance with a Mean Absolute Percentage Error (MAPE) of 0.76% and an R² score of 0.96, indicating strong predictive capability and generalization performance. The trained model was integrated into a Flask-based web application that enables users to retrieve land information, estimate land prices, and calculate land taxes in real time using Parcel Unique Identification (UPI).

The findings confirm that machine learning techniques can significantly improve valuation accuracy, consistency, transparency, and scalability in land administration systems. The proposed framework contributes toward the modernization of land governance in Rwanda by supporting intelligent and data-driven valuation processes.