A new hybrid fuzzy bio-inspired classifier for cancer detection using cuckoo optimization and hyper-planes

Abstract

This paper presents a novel hybrid fuzzy bio-inspired classification method for cancer detection, designed to address the challenges of classifying imbalanced medical data. The proposed method, named FCOA-HPC, integrates a new fuzzy draft of the Cuckoo Optimization Algorithm (FCOA) with the Hyper-Planes Classifier (HPC). The core innovation lies in using FCOA to dynamically optimize the weights of multiple separating hyper-planes within the HPC framework. A significant enhancement is the introduction of a Mamdani fuzzy inference system to intelligently adjust the Egg Laying Radius (ELR) parameter of the COA in real-time, based on the algorithm's convergence state and current best profit. This adaptive tuning mechanism allows for a more efficient and effective search of the solution space, balancing global exploration and local exploitation to find optimal hyper-plane configurations.

The research is grounded in a comprehensive technical review of existing cancer detection methods, systematically categorizing them and analyzing their structural advantages and disadvantages. This analysis identifies HPC as a superior classifier due to its linear time complexity, fast implementation, suitability for imbalanced data, and lack of need for prior knowledge. Concurrently, the review highlights the common challenges in tuning parameters for various meta-heuristic optimizers, motivating the design of the more adaptive FCOA. The proposed hybrid method was rigorously evaluated on four well-known UCI cancer datasets—WOBC, WDBC, Breast Cancer Coimbra, and Lung Cancer—using different numbers of hyper-planes (1 to 4) and various train-test splits (25%-75%, 50%-50%, 75%-25%).

The experimental results demonstrate the outstanding performance of the FCOA-HPC hybrid. It consistently achieved high classification accuracy across all datasets, with particularly strong results in scenarios with limited training data (e.g., 25% training). For instance, on the WDBC dataset, FCOA-HPC reached a total accuracy of 99.55%, and on WOBC, it achieved 99.39%. A key positive finding is the small gap between training and testing accuracy, indicating excellent generalization and robustness without overfitting. The method also showed faster convergence and lower error rates compared to the standard COA-HPC, validating the effectiveness of the fuzzy-based parameter adaptation. Furthermore, when compared with over 30 existing methods from the literature, FCOA-HPC matched or surpassed the performance of state-of-the-art techniques, often while offering the advantage of linear computational complexity, making it efficient for practical use.

In conclusion, this study makes a significant contribution by developing an accurate, efficient, and robust hybrid classifier for cancer detection. The integration of fuzzy logic into the bio-inspired optimizer successfully addresses a key parameter-tuning challenge, leading to a method that excels with imbalanced data and limited training samples. The FCOA-HPC framework proves to be a highly adaptable and powerful tool for medical diagnosis, with the potential to improve early and accurate cancer detection in clinical settings. Its design principles also open avenues for future research into integrating other adaptive optimization strategies for further enhancement.