On multilabel classification methods of incompletely labeled biomedical text data.

Area of Science:

• Machine Learning
• Data Science
• Bioinformatics

Background:

• Multilabel classification is challenged by incomplete training datasets where labels may be missing.
• Directly training classifiers on such incomplete data leads to ineffectual performance.
• The accuracy of classification is compromised when the completeness and correctness of labels are unknown.

Purpose of the Study:

• To enhance the performance of multilabel classification algorithms when faced with incompletely labeled training data.
• To introduce and evaluate a training set modification step prior to classifier training.
• To investigate the efficacy of weighted k-nearest neighbor (WkNN) and soft supervised learning (SoftSL) for modifying training sets.

Main Methods:

• Implemented two training set modification algorithms: weighted k-nearest neighbor (WkNN) and soft supervised learning (SoftSL).
• Both WkNN and SoftSL methods rely on similarity measurements between data vectors.
• Conducted experiments using Support Vector Machine (SVM) and Random Forest (RF) classifiers on both original and modified datasets.

Main Results:

• Experiments were performed on the AgingPortfolio (text) and Yeast (genetic) datasets.
• Both WkNN and SoftSL, as preprocessing steps, led to considerable performance improvements for SVM and RF classifiers.
• The training set modification approach demonstrated superior results compared to direct training on incomplete datasets.

Conclusions:

• Training set modification is a crucial step for improving multilabel classification accuracy with incomplete data.
• WkNN and SoftSL are effective methods for modifying training sets, enhancing classifier performance.
• The proposed approach is applicable to diverse data types, including text and genetic data.