19th INTERNATIONAL ISTANBUL CONGRESS ON LIFE, ENGINEERING, ARCHITECTURE, AND MATHEMATICAL SCIENCES
A NOVEL METAHEURISTIC APPROACH FOR HYPERPARAMETER TUNING IN MACHINE LEARNING MODELS
Yayıncı:
BZT Turan Publishing House
Machine learning algorithms are widely used across various fields, from customer churn prediction and sales forecasting to image detection. Most machine learning algorithms have hyperparameters that require tuning for optimal performance. Conventional tuning methods, such as trial and error or brute-force approaches, are highly time-consuming and, in some cases, impractical for complex problems due to their NP-hard nature. As a result, researchers and practitioners turn to metaheuristic methods for solving and optimizing machine learning algorithms. However, it remains uncertain which metaheuristic algorithm is best suited for a given problem, and while metaheuristics provide acceptable solutions, they do not guarantee optimal ones. This leaves an opportunity for improvement through the development of novel metaheuristic approaches. In this study, we propose a novel metaheuristic approach by hybridizing Ant Colony Optimization (ACO) with the Honey Badger Algorithm (HBA) to leverage the strengths of both algorithms, aiming to create more robust metaheuristic algorithm. We intend to test our hybrid method on various tasks, such as future demand forecasting and customer churn prediction, comparing its performance against well-known metaheuristic algorithms using statistical metrics like Mean Absolute Percentage Error (MAPE). Ultimately, our goal is to demonstrate that this novel hybrid approach can achieve improved accuracy and adaptability in diverse optimization tasks.
