An Efficient Filter-Based Feature Selection Model to Identify Significant Features from High-Dimensional Microarray Data

Subset feature selection for microarray data is intended to reduce the number of irrelevant features in order to extract significant features from the dataset. Simultaneously, choosing the appropriate features from the high-dimensional dataset may enhance the learning algorithm’s classification precision. Relief algorithms and their variants are successful attribute estimators. MultiSURF is the latest descendant of relief-based approaches; it estimates the feature by preserving target instance-centric neighborhood determination. However, the large number of redundant features can lead to the degraded performance of MultiSURF’s ability to select relevant features from the datasets. In order to select an informative feature from extremely redundant data, we suggest an innovative feature weighting algorithm called boundary margin relief (BMR). BMR’s main concept is to predict the feature weights through the measurement of the local hyperplane, which is typically used in I-Relief techniques. The weights of the features in the proposed methods are very robust in terms of removing redundant features. To show the efficiency of the method, comprehensive studies involving classification tests were conducted on benchmark microarray datasets by combining the suggested technique with conventional classifiers, including support vector machine, k-nearest neighbor, and Naive Bayes. Extensive studies have shown that the proposed method has three notable features: (1) elevated classification accuracy, (2) outstanding redundant robustness, and (3) good stability about different classification algorithms.