This study explores the efficacy of Bayesian hyperparameter optimization (BHO) to fine-tune machine learning algorithms specifically employed for breast cancer diagnosis as a case study of disease diagnosis using machine learning. BHO leverages a probabilistic model-based approach, utilizing Gaussian processes as surrogate models and the expected improvement acquisition function to intelligently navigate the hyperparameter space. Models used in this study, including support vector machines, random forests, gradient boosting machines, logistic regression, and K-nearest neighbor were optimized over parameters such as penalty coefficients, tree depths, learning rates, and architectural configurations. The objective is to maximize cross-validated accuracy using the Wisconsin Breast Cancer Diagnostic Dataset. Initial results were established through a set of random evaluations, followed by iterative refinements based on model predictions and performance feedback. The findings demonstrate that Bayesian hyperparameter optimization significantly improves the predictive performance of the models, suggesting that this approach is a viable and effective strategy for enhancing machine learning applications in breast cancer diagnosis. This study not only underscores the potential of Bayesian optimization in medical AI but also sets a precedent for its application in other areas of health informatics.

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Optimizing Machine Learning Models for Disease Diagnosis Using Bayesian Hyperparameter Optimization (BHO)

  • O. Olawale Awe,
  • Jeremiah M. Adepoju

摘要

This study explores the efficacy of Bayesian hyperparameter optimization (BHO) to fine-tune machine learning algorithms specifically employed for breast cancer diagnosis as a case study of disease diagnosis using machine learning. BHO leverages a probabilistic model-based approach, utilizing Gaussian processes as surrogate models and the expected improvement acquisition function to intelligently navigate the hyperparameter space. Models used in this study, including support vector machines, random forests, gradient boosting machines, logistic regression, and K-nearest neighbor were optimized over parameters such as penalty coefficients, tree depths, learning rates, and architectural configurations. The objective is to maximize cross-validated accuracy using the Wisconsin Breast Cancer Diagnostic Dataset. Initial results were established through a set of random evaluations, followed by iterative refinements based on model predictions and performance feedback. The findings demonstrate that Bayesian hyperparameter optimization significantly improves the predictive performance of the models, suggesting that this approach is a viable and effective strategy for enhancing machine learning applications in breast cancer diagnosis. This study not only underscores the potential of Bayesian optimization in medical AI but also sets a precedent for its application in other areas of health informatics.