Machine Learning-Based House Price Prediction via Whatsapp and Website Chatbot

Abstract

House property valuation in Masvingo rely on manual appraisals and comparative market analysis, which are often slow, inconsistent, and unable to adapt to rapid market changes. Recent advances in machine learning and digital valuation platforms have demonstrated improved estimation accuracy, though adoption in Masvingo remains limited. The aim of this study was to develop a machine learning-based model to accurately predict house prices in Masvingo Urban via WhatsApp and Website Chatbot. Primary and secondary data was collected from local real estate agencies using questionnaires, interviews, and document analysis. Data preprocessing and synthetic data generation was employed to address data sparsity. After augmenting the dataset, recalibration was performed to ensure the synthetic data matched the statistical characteristics of the original dataset and to reduce overfitting. Random Forest Regression, which outperformed other algorithms, had a Mean Absolute Error of 464.67 and R² of 0.8298. The algorithm predicted a high-density Masvingo property (500 m² stand, 200 m² building, seven rooms, standard amenities and standard materials) at 30,784.55. This study is essential for real estate professionals, buyers, and policymakers, providing standardised, fast, and data-driven property valuation. The study recommends that machine learning models offer a practical ICT solution for sustainable real estate valuation in Masvingo Urban.