SYNTHETIC DATA AS A FOUNDATION FOR A SELF-LEARNING RECOMMENDER SYSTEM IN BANKING: MODELING CLIENT BEHAVIOR AND MACROECONOMIC FACTORS

Askerov Zaur Khanakhmedovich

The development of modern intelligent recommender systems in the banking sector is constrained by limited access to real customer data due to legal and ethical concerns. This paper presents a methodological framework for building a self-learning recommender system based on the comprehensive generation of synthetic data. The system architecture includes a multi-agent model that simulates transactional, deposit, and investment behaviors of clients. Data generation incorporates language models, stochastic processes, and procedural simulations based on behavioral profiles. Special emphasis is placed on integrating macroeconomic indicators (exchange rates, commodity prices, interest rates) into the synthetic environment to recreate realistic scenarios of financial instability. A visual and statistical analysis of the generated dataset confirms its adequacy for training neural network models and reinforcement learning algorithms. The proposed approach ensures reproducibility, scalability, and data privacy in the development of AI-powered financial systems.

Keywords:synthetic data, recommender systems, reinforcement learning, multi-agent simulation, banking technology, client anomaly detection, macroeconomic indicators, data generation, behavioral modeling, intelligent systems.