L1-Regulated Feature Selection in Microarray Cancer Data and Classification Using Random Forest Tree

Microarray cancer data are characterized by high dimensionality, small sample size, noisy data, and an imbalanced number of samples among classes. To alleviate this challenge, several machine learning-oriented techniques are proposed by authors from several disciplines such as computer science, computational biology, statistics, and pattern recognition. In this work, we propose L1-regulated feature selection method and classification of microarray cancer data using Random Forest tree classifier. The experiment is conducted on eight standard microarray cancer datasets. We explore the learning curve of the model, which indicates the learning capability of the classifier from a different portion of the training samples. To overcome the overfitting problem, feature scaling is carried out before the actual training takes place and the learning curve is explored using fivefold cross-validation method during the actual training time. Comparative analysis is carried out with state-of-the-art work, and the proposed method outperforms many of the recently published works in the domain. Evaluation of the proposed method is carried out using several performance evaluation techniques such as classification accuracy, recall, precision, f-measure, area under the curve, and confusion matrix.