This project aims to use machine learning algorithms, specifically Convolutional Neural Networks (CNNs) and Multilayer Perceptrons (MLPs), to predict house prices based on images and numerical data. The goal is to determine the importance of photos in predicting property values by employing dimensional reduction and feature selection techniques.

The real estate industry heavily relies on visual appeal and perceived value when it comes to selling properties. However, accurately determining the true value of a property based solely on images can be challenging due to factors like lighting conditions, angles, and subjective perception. This project aims to address this challenge by leveraging machine learning algorithms to predict house prices using both numerical data (e.g., square footage) and image-based features (e.g., number of bedrooms). Additionally, the project will investigate the significance of photos in predicting property values by employing dimensional reduction and feature selection techniques.

This approach differs significantly from previous approaches through its modalities. A majority of real estate value predictions are used solely with numerical and categorical data about the real estate property information or locality information from geospatial data. However, this multimodal approach uses specifically visual images of the Southern California real estate properties as well as the aforementioned property information.

View our presentation here!