Real Estate Predictive Analytics Using Multi-Model Regression– Classification Pipeline with Recurrent Learning

Abstract

Real-estate valuation plays a crucial role in guiding buyers, sellers, and investors, yet traditional methods still rely heavily on manual inspection, subjective judgment, and limited comparable market data. Existing automated valuation approaches often treat the task purely as a regression problem, ignoring classification components and failing to handle imbalanced class distributions. Moreover, traditional modelling techniques struggle with non-linear patterns and high-dimensional features, resulting in limited generalization and inconsistent pricing outcomes. Conventionally, valuation is performed manually by comparing listings, interpreting property documents, and adjusting prices based on personal experience. The proposed system addresses these challenges by integrating a complete machine learning workflow consisting of preprocessing, Synthetic Minority Over-sampling Technique (SMOTE)–based class balancing, exploratory data analysis, and multi-model training tailored for both regression and classification tasks. In the regression module, the framework predicts numerical real-estate attributes such as property price measured in lakhs of rupees and built-up area measured in square feet (SqFT), ensuring accurate continuous-value estimation across diverse property types. For the classification module, the system categorizes directional property orientation into facing classes such as North (N), South (S), East (E), and West (W) which are crucial determinants in architectural design, ventilation, sunlight exposure, and market valuation. Three model families are utilized: Gradient Boosting Classification and Regression Trees (Gradient Boosting CART), Extreme Gradient Boosting CART (XGBoost-CART), and a hybrid Recurrent Learning framework such as Long Short-Term Memory along with Decision Tree (DT-CART) also known as LSTM–TreeNet architecture, enabling parallel processing of both regression and classification objectives.