SPLIT-FED LEARNING APPROACH FOR HOUSE PRICE PREDICTION WITH HETEROGENEOUS FEATURES

Date

2025-12-17

Author

Nikel, Ayşenur

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

130
views

0
downloads

Cite This

Federated Learning is an emerging approach for collaborative training of a common global model without disclosing raw data available in multiple clients. Previous studies mainly focus on two distinct types of heterogeneity in federated learning, namely device heterogeneity and model heterogeneity. While feature space heterogeneity presents a significant challenge in real-world applications, there are a limited number of works that address this issue. In this thesis, we address this challenge in the context of the house price prediction task. Our primary objective is to enable clients to collectively train a global model by utilizing all the features they have. To achieve this, we employed a Split-Fed method, where clients retain the local portion of the model and their data in a private manner while training the upper portion of the model collectively. Five city datasets have been used for evaluations. The experiment results have shown that our Split-Fed method outperforms individual city models trained by all available features for the same city, individual city models trained solely using the common features across all cities, a global model trained using the merged dataset with the common features, and a global model trained using the classical federated learning approach with only the common features. Consequently, we have shown that the proposed approach is very promising in addressing the challenge of feature heterogeneity in federated learning.

Subject Keywords

Federated Learning, Split Learning, Split-Fed Learning, Feature Space Heterogeneity

URI

https://hdl.handle.net/11511/118201

Collections

Graduate School of Informatics, Thesis