SMART BIDDING: A MACHINE LEARNING APPROACH TO ONLINE AUCTIONS WITH LSTM AND KALMAN FILTERS

Abstract [English]

The evolution of online auction systems has revolutionized the way products and services are bought and sold. Among various auction formats, the English auction model remains one of the most widely adopted due to its simplicity, scalability, and ability to accommodate a large number of participants in real-time. These systems, also known as e-Auctions or electronic auctions, have become increasingly prevalent across industrial and commercial sectors due to their efficiency, cost-effectiveness, and transparency.
To enhance the predictive capabilities and decision-making accuracy of these systems, advanced machine learning techniques are integrated into the auction process. Specifically, Long Short-Term Memory (LSTM) networks are employed to capture temporal bidding patterns and forecast price trends over time. LSTMs are particularly effective in handling sequential data and learning complex dependencies, making them suitable for dynamic, time-sensitive auction environments.
In addition, Kalman Filters are utilized to refine real-time predictions by continuously updating and correcting the forecasted price based on new observations. This hybrid approach—combining LSTM’s deep learning with Kalman filtering’s real-time estimation—enables more accurate and adaptive prediction of final auction prices, even in the presence of noisy or incomplete data.
The system is capable of handling diverse product categories, from hardware items with quantifiable features (e.g., memory size, processing speed) to "soft" goods such as jewelry, which rely on subjective attributes like color, texture, and design. By leveraging these intelligent algorithms, the auction system improves the reliability of price prediction, enhances bidder engagement, and supports better-informed decision-making for both buyers and sellers