In the past few years, Lopsidedness of class is the major challenge apart from other issues, which obtains much more debate in the field of big data research, while someone interacts with real-world data sources emerging from e.g. communication technologies, genetics, ecology, and forgery identification. Enhancing the reliability in order to detect their minority class is an incredibly significant problem for this sort of dataset. There is a procedure to resolve this problem, called feature selection. A framework of embedded feature selection on the basis of Weighted Gini Index (WGI) is built through this earlier research project, simultaneously WGI approach also has several restraints. A separating node can be simply modified within a decision tree algorithm. Also, when several features reach a particular frequency concurrently, it becomes complicated to discern which function can be chosen initially. In order to deal with this complication, the suggested method developed a Glowworm Swarm Optimization Algorithm (MSGSO) based on Modified Step size for an effective selection of features. The suggested method comprises the phases for the oversampling, feature selection, and classification. Earlier, the Synthetic Minority Oversampling TEchnique (SMOTE) is enforced on the input dataset. This is one among the oversampling strategies, which is being employed very often for clearing up the issue of lopsidedness, that produces artificial training samples for the minority class through linear initiation. Thereafter, the Modified Step size depending Glowworm Swarm Optimization Algorithm (MSGSO) is applied to choose an appropriate feature. Eventually, categorization is carried out through the Improved Regularization based Convolutional Neural Network (IRCNN). Contrasted to the current methods, the empirical findings of the suggested method indicate that it acquires a higher efficiency with regards to accuracy, precision, recall, and f-measure.
Volume 12 | Issue 4
Pages: 390-400
DOI: 10.5373/JARDCS/V12I4/20201453