Multi-class classification using Cuckoo Search based hybrid network

Classification plays an important role in various fields such as science, engineering, medicine and business. This paper proposes a cuckoo search based hybrid model i.e. Functional Link Neural Fuzzy Network named as CSFLNFN for classification of multi-class datasets. Both FLANN, as an efficient computational technique and fuzzy logic, as a basis of much inference system are combined to take the advantages from both the techniques. These two techniques are supplementary to each other in a way that one is helping other to overcome their limitations. The proposed CSFLNFN model uses FLANN to the consequent part of the fuzzy rules. The parameters of the models are optimized by the evolutionary algorithm, Cuckoo Search (CS). The CSFLNFN model is evaluated with one medical dataset, dermatology and three other frequently used multi-class datasets, wine, glass and iris. Further, to get more classification accuracy, Principal Component Analysis (PCA) has been used to extract the features from the datasets. Performance of the model is measured by number of measures like confusion matrix, accuracy, sensitivity, specificity, F-score, gmean and area under the receiver operating characteristic (ROC) curve. In this study, a comparison has been made between results before and after features extraction and it is seen that the classification accuracy increases with extracted features from the datasets. However the results demonstrate the superiority of the CSFLNFN compare to other models including CSMLP, CSFLANN, Naïve Bayesian and K-Nearest Neighbor irrespective of the feature extraction.