An Efficient Cancer Classification Model Using Microarray and High-Dimensional Data

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Cancer remains a leading cause of death globally.
• Microarray gene expression profiling is crucial for cancer diagnosis, prognosis, and treatment.
• High-dimensional, redundant gene expression data presents significant classification challenges.

Purpose of the Study:

• To develop a hybrid machine learning approach for effective cancer classification.
• To address the NP-Hard problem of gene expression profile classification.
• To reduce dimensionality and select informative genes from microarray data.

Main Methods:

• Utilized a hybrid model combining Pearson's correlation coefficient for feature selection/reduction.
• Employed a Decision Tree classifier for its interpretability and ease of use.
• Optimized the Decision Tree's maximum depth hyperparameter using Grid Search Cross-Validation (CV).

Main Results:

• Evaluated the model on seven standard microarray cancer datasets.
• Demonstrated significant reduction in the number of genes required for classification.
• Showcased enhanced classification accuracy, specificity, sensitivity, F1-score, and AUC.

Conclusions:

• The proposed hybrid strategy effectively reduces gene numbers and selects the most informative features.
• The model significantly improves cancer classification accuracy using gene expression data.
• This approach offers a robust method for analyzing complex genomic datasets in oncology.