A SCALABLE HYBRID MACHINE LEARNING FRAMEWORK FOR RENTAL PRICE CLASSIFICATION LEVERAGING SEMANTIC TEXTUAL EMBEDDINGS AND SENTIMENT-DRIVEN FEATURE ENGINEERING

Abstract

In today’s digital era, rental listing platforms host thousands of property ads, making it increasingly difficult for users to assess price categories and for agents to market properties effectively. While structured features like location, number of bedrooms, and square footage are widely used, they often fail to capture the full context provided in descriptions and amenities, which influence viewer interest and pricing perception. Traditional systems primarily rely on rulebased filters or manual interpretation, which are not only time-consuming but also inconsistent and incapable of adapting to the growing scale and diversity of listings. The limitations of these traditional approaches include poor handling of textual data, low scalability, and limited accuracy in predicting appropriate price tiers. This leads to mismatches in user expectations, reduced engagement, and inefficiencies in rental market operations. To overcome these challenges, there is a clear need for a system that intelligently leverages both structured and unstructured data to predict the rental price category more accurately. This research proposes a machine learning-based classification system that incorporates both numerical property features and natural language data (such as descriptions and amenities) using advanced preprocessing techniques and TF-IDF vectorization. Three models—Support Vector Classifier (SVC), KNearest Neighbors (KNN), and Light Gradient Boosting Machine (LGBM)—are implemented and compared. The LGBM model significantly outperforms others, achieving near-perfect accuracy and balanced performance across price categories. The system offers a scalable, accurate, and intelligent solution to predict rental price ranges, making it highly valuable for real estate platforms aiming to optimize pricing strategies and enhance user experience