Biased Random Forest For Dealing With the Class Imbalance Problem.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Class imbalance is a major challenge in machine learning, biasing algorithms towards majority classes and hindering accurate predictions.
• This issue is prevalent in real-world applications, complicating predictive analysis due to unequal class distributions.
• Existing methods often struggle with complex data factors, necessitating new approaches for effective minority class identification.

Purpose of the Study:

• To propose a novel ensemble-based method to effectively address the class imbalance problem.
• To shift the oversampling strategy from the data level to the algorithm level within ensemble methods.
• To enhance the representation of minority classes within the classification ensemble, specifically a random forest.

Main Methods:

• Developed a biased random forest algorithm that moves oversampling from data to algorithm level.
• Utilized the nearest neighbor algorithm to pinpoint critical areas within the dataset.
• Generated additional random trees for the random forest ensemble, focusing on these identified critical areas.

Main Results:

• The proposed biased random forest algorithm demonstrated significant effectiveness in handling class imbalance.
• The method successfully improved the recognition of the minority class during classification tasks.
• Results indicate a superior performance compared to standard approaches in imbalanced datasets.

Conclusions:

• The novel ensemble method provides an effective solution for the persistent class imbalance problem.
• Oversampling the classification ensemble, rather than the data itself, is a viable and effective strategy.
• The biased random forest algorithm offers a promising direction for improving predictive accuracy in imbalanced learning scenarios.