Rental Price Prediction using Machine Learning - a French Case-study - Comparative Thesis Results

Aug 3, 2026

Upload by Quinn

Real estate represents a major store of wealth and housing prices serve as key economic indicators, making accurate rent-price prediction a significant challenge. This thesis examines how Artificial Intelligence and Machine Learning can automate pricing through Automated Valuation Models (AVMs) that estimate property value from data. It compares a global AVM Rent model for France with city-specific local models for Paris, Lyon, Toulouse, Marseille, Nice, and Montpellier. Results indicate higher accuracy with local models, and XGBoost achieves the best performance with an MDAPE below 8.4% across all six cities.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages78

Reading Time3 h 17 min

AccessFree to download

Introduction

Context

State of the Art and Related Work

Theoretical Background

Machine Learning

Supervised Learning

Model Evaluation

Evaluation Metrics

Regression Metrics

Train, Validation, Test splits

Cross-Validation

Hyperparameters Tuning

Method

Problem Formulation

Models

GAM

Ensemble Models

Elasticnet

Experiments

Datasets

Preprocessing

Model Building

Testing Local Models

Testing Local Models on different Cities

Local Models vs. PriceHubble Global Model

Global XGBoost

Limitations of the Local Models

Conclusion

Future Works

Rental Price Prediction using Machine Learning: a French Case-study Master’s Degree of Science Thesis in Mathematical Engineering-Statistical Learning Author: Camilla Maria Caroni Student ID: 971053 Advisor: Francesco TROVÒ PhD Co-advisors: Paolo BIGHIGNOLI Academic Year: 2021-22 i Abstract Real Estate is the world’s most significant store of wealth and it has been continuously growing in the last few years. The real estate market has a close relationship with us: it plays a very important role in economic development and people’s fundamental need. Housing price, in particular, is an important reflection of the economy and an indicator of the healthy and stable development of Real Estate. So, accurately predicting real estate prices poses a significant challenge to the players present in this market. Historically considered relatively conservative and characterized by qualitative approaches and processes, the real estate industry has been revolutionized in recent years by introducing Artificial Intelligence (AI) and Machine Learning (ML) . One of the most beneficial applications of ML in Real Estate is the automation of the pricing process via the so-called Automated Valuation Models (AVMs) . AVMs are ML models trained to predict the value of a particular property. This thesis aims at comparing the performances of a global AVM Rent model for France with the performances of some other models which are instead local, i.e. , built specifically for each given city among the following ones: Paris, Lyon, Toulouse, Marseille, Nice, and Montpellier. Results showed that a larger prediction accuracy can be achieved by exploiting local models rather than a global one. In particular, among all the examined models, XGBoost provides the most promising option for the task, with an MDAPE lower than 8 .4% in all 6 cities analyzed. Keywords: real estate market, machine learning, automated valuation models, house rent pricing. Sommario Il mercato immobiliare è la piu grande riserva di ricchezza al mondo e negli ultimi anni è stato caratterizzato da una continua espansione. Il mercato immobiliare stringe con noi un rapporto molto stretto: gioca infatti un ruolo fondamentale all’interno dello sviluppo economico e dei bisogni primari delle persone. Il prezzo delle case, in particolare, è unimportante riflesso dell’economia, nonchè un indicatore di uno sviluppo sano e solido dellostesso mercato immobiliare. Predire in maniera accurata i prezzi dei beni immobiliaricostituisce quindi una sfida di grande rilenvanza per gli attori presenti in questo mercato. Storicamente considerata relativamente conservativa e caratterizzata da approcci e processi qualitativi, l’industria immobiliare è stata recentemente rivoluzionata dall’introduzione dell’Intelligenza Artificiale (IA) e del Machine Learning (ML) . Una delle applicazioni più vantaggiose del ML in ambito immobiliare è l’automazione dei processi di pricing attraverso i cosiddetti Automated Valuation Models (AVMs) . Gli AVMs sono modelli di ML addestrati a predire il valore di un particolare bene immobiliare. Questa tesi mira a confrontare la performance di un modello AVM globale per il prezzo di affitto di case e appartamenti in Francia con le performances di alcuni altri modelli che sono invece locali, i.e. , costruiti specificamente per ciascuna delle seguenti città: Parigi, Lione, Tolosa, Marsiglia, Nizza e Montpellier. I risultati ottenuti mostrano che predizioni più accurate possono essere ottenute sfruttando i modelli locali piuttosto che quello globale. In particolare, tra tutti i modelli esaminati, XGBoost costituisce l’opzione più promettente per il task, con un MDAPE inferiore all’8 .4% in tutte e 6 le città analizzate. Keywords: mercato immobiliare, machine learning, modello di valutazione automatizzato, house rent pricing. v Contents Abstract i Sommario iii Contents v Introduction 1 1 Context 5 1.1 PriceHubble ..............................

FAQs

What problem does the thesis address?

The thesis tackles accurate rental price prediction in the real estate market, where housing prices strongly reflect economic conditions.

How are global and local AVM models compared?

It compares a global AVM Rent model for France against local models built specifically for each analyzed city: Paris, Lyon, Toulouse, Marseille, Nice, and Montpellier.

Which model performs best and what metric is reported?

XGBoost shows the most promising results, achieving an MDAPE lower than 8.4% in all six cities.

Aug 3, 2026

Upload by Quinn