Clustering analysis for classifying fake real estate listings

Area of Science:

• Data Science
• Machine Learning
• Real Estate Analytics

Background:

• The proliferation of online real estate platforms has led to an increase in fraudulent listings, causing significant user detriment.
• Existing fraud detection methods in real estate are limited, particularly for distinguishing fake listings on rental and sale platforms.

Purpose of the Study:

• To apply clustering analysis for classifying real estate listings as genuine or fake.
• To enhance the accuracy of fake listing detection using optimized K-means clustering.

Main Methods:

• Data pre-processing and feature engineering on expert-curated datasets.
• K-means clustering algorithm with parameter optimization using Silhouette coefficient, Calinski-Harabasz index, and Davies-Bouldin index.
• Evaluation of clustering performance using Random Forest and Decision Tree classifiers, with Camberra distance metric identified as optimal.

Main Results:

• Optimized K-means clustering successfully differentiated fake from genuine real estate listings.
• The clustering approach improved the accuracy of the Random Forest classification model by 96%.
• The study generated balanced datasets comprising distinct fake and genuine property listing clusters.

Conclusions:

• Clustering analysis, specifically optimized K-means, offers a robust solution for identifying fake real estate listings.
• This approach enhances the reliability of online property platforms and aids future deep learning model development.
• The findings contribute to a more secure and trustworthy real estate transaction environment.