Can AI predict your property price based on images?
Cosmin Cojocaru is activating as a Data Science Lead at DB Global Technology, Deutsche Bank’s technology centre in Central and Eastern Europe and is currently leading cross-functional international teams in developing and deploying risk models for Anti-Fraud, Bribery and Corruption area. He is also a member of DB’s Data Science Center of Expertise, being involved in leading the development of knowledge and expertise across the bank by building learning pathways and knowledge assets to support learning, best practices and career journeys in the field.
As a certified trainer, Cosmin is currently an Associate Lecturer with the Romanian Banking Institute, where he teaches several Data Science courses and he is co-founder of aiacademy.ro, which focuses on delivering courses and training in the AI field.
Cosmin’s experience is focused on developing ML solutions for AML/Compliance (classification problems, graph algorithms), Credit Risk (classification and regression problems), Corporate Sales (clustering, classification, visualizations), Collateral Management and Process Efficiency (OCR and natural language processing).
In this talk we will try to predict prices for residential properties having as input both structured data related to the property and also a set of images for each property.
We will develop a baseline model based on the structured data and we will then compare it with a Convolutional Neural Net that we will develop based only on images, to see which model has a more predictive power.
We will then combine the two models to evaluate if the baseline model combined with the information gain from the CNN improves the overall predictions of the consolidated model.