Classification of Biodegradable Substances Using Balanced Random Trees and Boosted C5.0 Decision Trees.

Area of Science:

• Environmental Science
• Cheminformatics
• Machine Learning

Background:

• Non-degradable substances pose environmental and health risks.
• Predicting biodegradability without expensive experiments is essential.
• Quantitative Structure-Activity Relationship (QSAR) models offer solutions but struggle with unbalanced data.

Purpose of the Study:

• To evaluate balanced random trees (RTs) and boosted C5.0 decision trees (DTs) for QSAR modeling of ready biodegradation.
• To assess the models' effectiveness in handling unbalanced molecular descriptor datasets.
• To compare their performance against existing classification methods.

Main Methods:

• Utilized a two-dimensional molecular descriptor dataset from the UCI machine learning repository.
• Implemented balanced RTs using balanced bootstrap samples.
• Applied cost-sensitive learning for boosted C5.0 DT modeling.
• Ranked molecular descriptors by contribution to classification.

Main Results:

• The proposed balanced RTs and boosted C5.0 DT models demonstrated superior performance compared to Support Vector Machine (SVM), K-nearest neighbors (KNN), and Discrimination Analysis (DA).
• Detailed classification metrics including accuracy, sensitivity, specificity, precision, F1 score, and ROC/AUROC were analyzed.
• Molecular descriptors were successfully ranked based on their importance in the classification process.

Conclusions:

• Balanced RTs and boosted C5.0 DTs are effective QSAR models for classifying ready biodegradation, especially with unbalanced data.
• These models offer improved accuracy and generalization capabilities for environmental risk assessment.
• The findings provide a more efficient approach to predicting substance biodegradability.