A Highly Discriminative Hybrid Feature Selection Algorithm for Cancer Diagnosis

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Oncology

Background:

• Cancer is characterized by uncontrolled cell proliferation, necessitating advanced diagnostic tools.
• Big data analytics offers potential for improved cancer diagnosis and understanding.
• Existing diagnostic methods may lack efficiency or accuracy in identifying specific cancer types.

Purpose of the Study:

• To propose a novel machine learning (ML) algorithm for diagnosing various cancer types from big data.
• To develop a two-stage hybrid feature selection method to identify optimal cancer-related features.
• To evaluate the algorithm's diagnostic accuracy and efficiency compared to existing methods.

Main Methods:

• A two-stage hybrid feature selection process combining filter-based methods (chi-squared, F-statistic, mutual information) and a modified wrapper-based sequential forward selection.
• Utilized machine learning classifiers including support vector machine (SVM), decision tree (DT), random forest (RF), and K-nearest neighbor (KNN).
• Validated on four cancer microarray datasets using 10-fold cross-validation and hyperparameter tuning.

Main Results:

• Achieved 100% diagnostic accuracy for leukemia using SVM and KNN with only 5 features.
• Attained 100% accuracy for ovarian cancer and small round blue cell tumor (SRBCT) using SVM with 6 and 8 features, respectively.
• Reached 99.57% accuracy for lung cancer using SVM with 19 features, outperforming other algorithms in feature selection and accuracy.

Conclusions:

• The proposed ML algorithm demonstrates superior performance in cancer diagnosis.
• The hybrid feature selection method effectively reduces dimensionality while maintaining high accuracy.
• This approach offers a promising tool for efficient and accurate cancer diagnostics in big data environments.