Ivan Lorencin1*, Nikola Tanković1, Ariana Lorencin2, Matko Glučina3
1 Juraj Dobrila University of Pula, Faculty of Informatics, Pula, Croatia
2 Department of Gynecology and Obstetrics, General Hospital Pula, Pula, Croatia
3 Istrian University of Applied Sciences, Pula, Croatia
ivan.lorencin [at] unipu.hr
Abstract
In this study, we explore the application of Kolmogorov-Arnold Networks (KAN) in the field of cervical cancer diagnostics. Cervical cancer, a major health concern worldwide, requires efficient and accurate diagnostic methods for early detection and treatment. Traditional machine learning approaches, such as multilayer perceptron (MLP) and K-nearest neighbors (KNN), have shown high classification performance but often at the cost of complex and resource-intensive architectures. In contrast, KAN offers a simpler yet effective alternative.
Utilizing a publicly available dataset of cervical cancer data, which includes 859 samples with 36 input attributes and diagnostic outputs defined as Hinselmann, Schiller, cytology, and biopsy, we implemented KAN for classification. Given the significant class imbalance in the dataset, we also applied various class balancing techniques.
Our results indicate that KAN can achieve high classification performance with mean area under the receiver operating characteristic curve (AUC) and mean Matthew’s correlation coefficient (MCC) scores comparable to those obtained with more complex architectures. Specifically, the KAN models demonstrated robust diagnostic capabilities, achieving AUC and MCC scores above 0.9.
The simplicity of KAN architectures, combined with their strong performance metrics, underscores their potential as a practical tool in medical diagnostics. These findings suggest that Kolmogorov-Arnold Networks could be effectively utilized for cervical cancer screening and diagnosis, providing a balance of high accuracy, robustness, and reduced computational complexity. This approach could facilitate more accessible and efficient diagnostic processes, particularly in resource-limited settings.
Keywords: cervical cancer, class balancing techniques, Kolmogorov-Arnolds network, preventive screening,