Unsupervised Hierarchical Classification Approach for Imprecise Data in the Breast Cancer Detection

Area of Science:

• Statistics
• Machine Learning
• Medical Informatics

Background:

• Statistical methods for classification with imprecise data, particularly interval-valued data, are an active research area.
• Interval-valued data presents unique challenges in classification tasks due to inherent uncertainty.

Purpose of the Study:

• To introduce a novel unsupervised hierarchical classification method (HC-ID) for multivariate interval-valued data.
• To apply the HC-ID method for improved breast cancer diagnosis, distinguishing between benign and malignant masses.

Main Methods:

• Developed and applied a hierarchical classification for imprecise multivariate data (HC-ID).
• Compared the performance of HC-ID against a conventional hierarchical classification (HC) approach.
• Utilized real-world data for breast cancer diagnosis to evaluate the methods.

Main Results:

• The HC-ID procedure demonstrated superior performance over the conventional HC method.
• HC-ID achieved higher accuracy (0.80) compared to HC (0.66).
• HC-ID showed significantly improved sensitivity (0.61) versus HC (0.08), reducing false negatives.

Conclusions:

• The proposed HC-ID method offers a more accurate and sensitive approach for breast cancer diagnosis using interval-valued data.
• HC-ID effectively mitigates the impact of data variability and uncertainty, leading to more reliable diagnostic outcomes.
• The conventional HC method struggles with high data variability, resulting in a notable rate of false-negative diagnoses.