Ensemble Text-Visual Anomaly Detection for Authentic Real Estate Listings

Abstract

The rapid expansion of digital platforms in the real estate sector has intensified concerns regarding the authenticity of online property listings. Fraudulent advertisements not only result in financial losses for users but also erode trust in property platforms. Traditional verification methods, which rely on manual moderation or simple rule-based techniques such as keyword filtering, price comparison, and duplicate detection, are often inefficient and incapable of identifying complex fraud patterns. As the volume of listings grows, these conventional approaches struggle to ensure data reliability and platform security. To address these limitations, this study proposes an intelligent hybrid framework called ExtraNet for automated classification of real estate listings. The model integrates an Extra Trees Classifier (ETC) with an Artificial Neural Network (ANN) to leverage the strengths of both ensemble and deep learning techniques. The system processes structured tabular data containing key property attributes, including price, location, area size, number of rooms, amenities, and listing descriptions. A comprehensive preprocessing pipeline is applied to handle missing values, encode categorical variables, and normalize numerical features, ensuring improved model performance. For comparative analysis, baseline models such as Logistic Regression (LR), Nearest Centroid (NC) and K-Nearest Neighbors (KNN) are also implemented. The proposed solution is deployed as a user-friendly desktop application using a Tkinter-based graphical interface with (My Stretched Query Language) MySQL database integration for efficient data handling. Experimental results demonstrate that the ExtraNet model achieves a high classification accuracy of 99.61%, significantly outperforming traditional models. This approach provides a scalable and effective solution for detecting fraudulent listings and enhancing trust in digital real estate platforms.