Hybrid k -Nearest Neighbor Classifier.

Area of Science:

• Machine Learning
• Data Mining
• Pattern Recognition

Background:

• Conventional k-nearest neighbor (KNN) classification struggles with sparse, imbalanced, and noisy datasets.
• Existing KNN methods often fail to effectively utilize both local and global data characteristics.

Purpose of the Study:

• To address the limitations of conventional KNN on specialized datasets.
• To propose a robust classification approach for high-dimensional, noisy data.

Main Methods:

• A hybrid KNN (HBKNN) approach integrating local and global information was developed.
• A random subspace ensemble framework based on HBKNN (RS-HBKNN) was designed for noisy, high-dimensional datasets.
• Nonparametric tests were used for comparative analysis against other classification methods.

Main Results:

• The RS-HBKNN classifier demonstrated strong performance on real-world datasets.
• Experiments showed RS-HBKNN outperforms several state-of-the-art classification approaches.
• The method effectively handles issues like data sparsity, imbalance, and noise.

Conclusions:

• The proposed RS-HBKNN classifier offers a significant improvement over traditional KNN methods.
• This approach provides a robust solution for classification tasks involving complex, real-world datasets.
• RS-HBKNN is a promising advancement in machine learning for handling challenging data characteristics.