# RECORD c2b6a75d94d2ce27d185f96ffbc7fa5ee377ffff # TITLE: AN INTERPRETABLE ANALYSIS OF RESIDENTIAL APARTMENT PRICES IN TASHKENT USING LINEAR REGRESSION # YEAR: 2025.0 VENUE: III-an’anaviy xalqaro ilmiy-amaliy konferensiya materiallari DOI: 10.65164/zrcja795 # AUTHORS: Abdurasul Bobonazarov # PAGES: 6 CANDIDATE TABLE PAGES (1-based): [] # ABSTRACT: Accurate understanding of residential housing price formation is essential for urban planning, real estate market transparency, and informed decision-making by buyers and sellers. In recent years, online real estate platforms have emerged as valuable sources of large-scale market data; however, empirical studies based on local housing data in Uzbekistan remain limited. This paper presents an interpretable analysis of residential apartment prices in Tashkent using multiple linear regression applied to real listing data collected from an online real estate platform. ===== PAGE 1 (score 8, numeric-paren 0, table-words 0, var-words 8) ===== 433 AN INTERPRETABLE ANALYSIS OF RESIDENTIAL APARTMENT PRICES IN TASHKENT USING LINEAR REGRESSION Abdurasul Bobonazarov, Khamidulla Khabibullaev Turin Polytechnic University in Tashkent e-mail: a.bobonazarov@polito.uz Abstract: Accurate understanding of residential housing price formation is essential for urban planning, real estate market transparency, and informed decision-making by buyers and sellers. In recent years, online real estate platforms have emerged as valuable sources of large-scale market data; however, empirical studies based on local housing data in Uzbekistan remain limited. This paper presents an interpretable analysis of residential apartment prices in Tashkent using multiple linear regression applied to real listing data collected from an online real estate platform. The dataset consists of 7,421 apartment listings and includes structural characteristics (apartment size, number of rooms, floor level, and total building floors) as well as geographic attributes represented by latitude and longitude. A systematic preprocessing pipeline involving outlier removal and train–test splitting is employed. Several linear models, including ordinary least squares, Ridge, Lasso, and Elastic Net regression, are evaluated and compared using standard performance metrics. Experimental results show that all linear variants achieve comparable performance, with an R² value of approximately 0.67 on the test set, indicating that a substantial proportion of price variability can be explained using linear relationships. The analysis highlights the dominant role of spatial location and apartment size in price formation, while regression coefficients provide clear interpretability of individual feature effects. Residual analysis confirms the absence of systematic prediction bias, with increased variability observed for higher-priced apartments. Overall, the findings demonstrate that simple and interpretable linear regression models provide a robust baseline for residential price analysis in the Tashkent housing market. The study offers practical insights for market participants and lays the groundwork for future extensions incorporating nonlinear and spatially explicit modeling approaches. Keywords: Residential property prices; Linear regression; Housing market analysis; Spatial factors; Real estate listings; Data-driven analysis; Tashkent housing market. INTRODUCTION Residential housing markets play a critical role in urban economic development, influencing household welfare, investment decisions, and city planning strategies. Understanding the factors that determine apartment prices is therefore essential for policymakers, real estate professionals, and individual buyers. In recent years, the rapid growth of online real estate platforms has enabled access to large volumes of market data, creating new opportunities for data-driven analysis of housing price formation. A common and well-established approach to housing price analysis is the use of hedonic pricing models, where property prices are expressed as a function of structural, locational, and environmental attributes. Among these approaches, multiple linear regression remains widely used due to its transparency, ease of interpretation, and solid theoretical foundation. Several studies have demonstrated that linear regression models can effectively capture key price determinants such as apartment size, number of rooms, building characteristics, and location variables [1-3]. Despite the growing popularity of complex machine learning and deep learning models, linear regression continues to serve as a strong baseline for housing price analysis, particularly when interpretability is a primary objective [4,5]. Recent literature has increasingly explored advanced regression techniques and machine learning models for housing price prediction, including regularized linear models, ensemble methods,