Federated Learning-Based Knowledge Transfer for Multi-Sector Stock Forecasting

Supervisor: Turgud Valiyev

Author: N/A

Abstract

The increasing complexity and heterogeneity of global financial data limit the scalability of centralized learning systems. This research proposes a Federated Learning (FL)-based knowledge transfer framework to enable adaptive model sharing in stock datasets. The framework investigates how cross-client knowledge transfer can improve prediction accuracy compared to individual machine learning (ML) models and how it can positively affect low- data or emerging companies, while preserving data privacy. Through multi-cluster FL architectures and model-agnostic transfer mechanisms, the study explores how information learned in one cluster can be effectively transferred to another via shared global representations. The proposed system aims to establish a foundation for continual, cross-domain, and privacy- preserving financial intelligence, advancing both the theory and application of federated learning in economics and finance.

This study is unique because of the introduction of a knowledge-transferable federated architecture that dynamically shares learned patterns between heterogeneous stock clusters. It bridges the gap between local specialization and global generalization in financial prediction. The framework will employ Machine learning and deep learning models such as LSTM, Random Forest, Linear Regression, Elastic Net, and LightGBM to benchmark predictive performance. Three key metrics will be used to evaluate models which are R², RMSE, and MAE. The companies will be selected across multiple sectors based on daily stock data from 2000 to 2025.

The research aims to strengthen the link between AI-driven predictive analytics and investment decision-making by providing more reliable, data-efficient, and adaptive forecasting tools for investors. The study contributes to the broader goal of improving financial stability and sustainable investment strategies in the era of intelligent finance by demonstrating how federated knowledge transfer enhances market understanding and minimizes risk.

Keywords:

Federated Learning, Knowledge Transfer, Stock Market Forecasting, Financial Prediction, Investment Decision-Making