A novel deep machine learning algorithm with dimensionality and size reduction approaches for feature elimination:...

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Thyroid cancer incidence is rising despite advancements in diagnostic tools.
• Current thyroid cancer diagnosis lacks a standardized procedure, leading to varied testing and complex datasets.
• The resulting multi-dimensional big data with sparse, randomly distributed missing values pose significant challenges for machine learning algorithms.

Purpose of the Study:

• To develop an accurate and computationally efficient deep learning algorithm for thyroid cancer diagnosis.
• To address data singularity issues caused by randomly distributed missing data.
• To enhance machine learning model performance by reducing data dimensionality and sample redundancy.

Main Methods:

• Developed techniques to handle randomly distributed missing data.
• Implemented dimensionality reduction using inner and target similarity approaches to select informative datasets.
• Utilized hierarchical clustering for size reduction by eliminating redundant data samples.
• Trained and validated four machine learning algorithms on unseen data.

Main Results:

• Achieved 100% training accuracy.
• Attained 83% testing accuracy on unseen data, demonstrating robustness and generalization.
• Evaluated the computational time efficiency of the algorithms under uniform conditions.

Conclusions:

• The proposed deep learning approach effectively addresses challenges in thyroid cancer big data analysis.
• The developed methods for handling missing data and reducing dimensionality improve diagnostic accuracy.
• The algorithm demonstrates high accuracy and computational efficiency, offering a promising tool for thyroid cancer diagnosis.