FLEX-SMOTE: Synthetic over-sampling technique that flexibly adjusts to different minority class distributions

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Class imbalance poses a significant challenge in predictive modeling, particularly affecting the accuracy of minority class predictions.
• Existing synthetic minority over-sampling techniques (SMOTEs) often struggle with diverse dataset distributions, making it difficult to select the optimal method.
• The performance of SMOTEs can be limited as some focus on class borders while others target class cores.

Purpose of the Study:

• To introduce FLEX-SMOTE, a novel and flexible synthetic minority over-sampling technique designed to overcome the limitations of existing methods.
• To develop an adaptable SMOTE approach that can effectively handle varied and non-obvious class distributions in datasets.
• To enhance the predictive performance for minority classes in imbalanced datasets.

Main Methods:

• Proposing FLEX-SMOTE, a new technique for synthetic minority over-sampling.
• Utilizing a density function to characterize and understand minority class distributions.
• Selecting an over-sampling region based on the specific characteristics of minority classes within a dataset.

Main Results:

• Experimental results demonstrate that FLEX-SMOTE significantly improves the predictive performance on minority classes.
• The proposed method shows effectiveness across various dataset distributions, unlike traditional SMOTEs.
• FLEX-SMOTE's adaptive region selection based on density functions proves beneficial for imbalanced learning.

Conclusions:

• FLEX-SMOTE offers a flexible and effective solution for addressing class imbalance problems in machine learning.
• The technique's ability to adapt to different data distributions makes it a versatile tool for improving minority class prediction.
• This approach provides a valuable advancement in the field of imbalanced learning and over-sampling techniques.