Gene expression cancer classification using modified K-Nearest Neighbors technique.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Genomics

Background:

• Gene expression microarray classification is vital for cancer prediction and diagnosis.
• High-dimensional gene expression data (thousands of genes, dozens of samples) present classification challenges.
• Machine learning offers powerful tools for building accurate classification models.

Purpose of the Study:

• To introduce a novel classification technique, Modified k-nearest neighbor (MKNN), for gene expression data.
• To enhance the performance of the k-nearest neighbor (KNN) algorithm through robust neighbor selection.
• To evaluate MKNN's effectiveness in two scenarios: smallest modified KNN (SMKNN) and largest modified KNN (LMKNN).

Main Methods:

• Developed the Modified k-nearest neighbor (MKNN) algorithm with a new weighting strategy for robust neighbor selection.
• Applied MKNN in SMKNN and LMKNN variations to improve KNN performance.
• Conducted experiments on six diverse gene expression datasets.

Main Results:

• MKNN, in both SMKNN and LMKNN forms, demonstrated superior performance compared to traditional and recent classification methods.
• MKNN achieved higher classification accuracy, precision, and recall than KNN, weighted KNN, Support Vector Machine (SVM), fuzzy SVM, and Brain Emotional Learning (BEL).
• MKNN exhibited significantly reduced testing time compared to KNN and weighted KNN.

Conclusions:

• MKNN is an effective and efficient technique for gene expression data classification.
• The proposed weighting strategy enhances the robustness and performance of KNN.
• MKNN offers a promising approach for improving cancer prediction and diagnosis systems using gene expression data.