IG-ANGO: a novel ensemble learning algorithm for breast cancer prediction using genomic data

• Published in: Evolving systems Vol. 15; no. 6; pp. 2399 - 2418
• Main Authors: Shalini, M., Radhika, S.
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.12.2024; Springer Nature B.V
• Subjects: Accuracy; Adaptive algorithms; Algorithms; Artificial Intelligence; Breast cancer; Classification; Complex Systems; Complexity; Design optimization; Engineering; Ensemble learning; Feature selection; Females; Gene expression; Genomics; Machine learning; Medical prognosis; Older people; Original Paper; Performance prediction; Tumors; Womens health; United States > US; Greylag Goose Optimization; Northern Goshawk Optimization; Ensemble classifiers; Genomic data; Breast cancer; Information Gain; Detection; Crossover
• ISSN: 1868-6478; 1868-6486
• DOI: 10.1007/s12530-024-09619-z

Breast cancer (BC) stands as the most prevalent form of cancer among women and is one of the primary causes of mortality worldwide. With recent advances in genomic research, precision medicine is now being used to treat BC. This study proposes a new Adaptive gene selection-enabled fine-tuned ensemble learning algorithm for BC prediction using Genomic data. We design an information gain-adaptive Northern Goshawk Optimization (IG-ANGO) based feature selection algorithm that uses inertia weights to improve the algorithm's search efficiency and accelerate convergence to optimal solutions. In addition, the Information Gain (IG) based ANGO method, uses ranking data to enhance the quality of selecting key features. The algorithm then employs the Crossover Greylag Goose Optimization (CGGO) algorithm to randomly generate individuals to recommend ideal solutions (i.e. hyperparameter configurations) to the ensemble model. The goal of an ensemble-based classification method is to generate a method with superior predictive capability. The final prediction result is determined by weighted voting based on the prediction probability results. Our experiments demonstrate that our method enhances the classifier's performance compared to other existing methods. The result obtained by our model demonstrates the highest classification accuracy at 99.84%. The proposed approach thus opens the possibility of new chances in BC prediction with Genomic data.