Genetic hyperparameter optimization with Modified Scalable-Neighbourhood Component Analysis for breast cancer...

Area of Science:

• Medical Imaging and Diagnostics
• Machine Learning in Healthcare
• Computational Biology

Background:

• Early breast cancer detection is crucial for improving patient survival rates and treatment efficacy.
• Traditional diagnostic methods can be time-consuming, and existing data mining techniques often lack optimal prediction accuracy.
• Parametric classifiers struggle with open-set scenarios and generalizing to new data with limited instances.

Purpose of the Study:

• To develop and implement a non-parametric strategy for breast cancer detection by optimizing feature embeddings.
• To enhance the prediction rate and scalability of breast cancer classification models.
• To address the limitations of conventional parametric classifiers in handling diverse and evolving datasets.

Main Methods:

• Utilized Deep Convolutional Neural Networks (Deep CNN) and Inception V3 for visual feature learning.
• Employed Neighbourhood Component Analysis (NCA) criteria for preserving neighborhood outlines in semantic space.
• Proposed Modified Scalable-Neighbourhood Component Analysis (MS-NCA) for non-linear feature fusion and optimized distance learning.
• Implemented Genetic-Hyper-parameter Optimization (G-HPO) to fine-tune XGBoost, Naïve Bayes, and Random Forest models.

Main Results:

• The proposed MS-NCA method enhances feature fusion by optimizing distance-learning objectives, enabling direct computation of inner feature products.
• G-HPO successfully determined optimized hyperparameters for XGBoost, Naïve Bayes, and Random Forest models.
• Analytical results confirmed significant improvements in the classification rate for identifying normal and affected breast cancer cases.

Conclusions:

• The non-parametric approach focusing on feature embedding optimization offers a scalable and effective alternative to traditional methods.
• The integration of MS-NCA and G-HPO provides a robust framework for accurate breast cancer prediction.
• This research contributes to advancing machine learning applications in early disease detection and diagnosis.