Prototype Selection Method Based on the Rivality and Reliability Indexes for the Improvement of the Classification...

Area of Science:

• Computational chemistry and cheminformatics
• Machine learning and artificial intelligence
• Data mining and knowledge discovery

Background:

• Prototype selection is crucial in machine learning for improving model performance and interpretability.
• In Quantitative Structure-Activity Relationship (QSAR) studies, effective data preprocessing enhances model accuracy and algorithm efficiency.
• Existing methods may not achieve significant data reduction without compromising model performance.

Purpose of the Study:

• To propose an efficient prototype selection method for QSAR classification model preprocessing.
• To demonstrate the method's ability to reduce training dataset size significantly.
• To validate the maintenance or improvement of classification accuracy and predictive performance on external datasets.

Main Methods:

• Development of a novel prototype selection algorithm.
• Integration of the method into the preprocessing stage of QSAR classification model construction.
• Validation using 40 benchmark datasets with varying sizes and balancing ratios.

Main Results:

• Achieved substantial reduction rates in training set cardinality.
• Maintained or increased the accuracy of QSAR classification models.
• Demonstrated preserved or improved prediction accuracy for external molecules.

Conclusions:

• The proposed prototype selection method is efficient for QSAR classification.
• The method effectively reduces data size while enhancing model accuracy and predictive power.
• The technique shows broad applicability across diverse datasets.