Techniques (Reference number is mentioned) | Segmentation | Feature extraction | Features | Classification | Advantages | Limitations | Accuracy |
[16] | Pixel based-Region growing | color and textural features | Extracted features are mean, standard deviation, variation, skewness, angular second moment, contrast, correlation, the sum of variance, inverse difference moment, sum average, sum variance, sum entropy, entropy, difference variance, difference entropy, information measures of correlation, and maximal correlation coefficient | SVM + kNN | The performance is high with the fusion of the SVM-kNN classifier. | Dataset is not standard and the feature extraction method is not defined. | 98% |
[20] | - | Rough Set-Based Feature Selection | Features are erythema, scaling, borders, itching, koebner phenomenon, polygonal and follicular papules, oral mucosal involvement, knee-elbow, and scalp involvement, family history, age, etc. | SVM + KNN + MLP | System accuracy is robust with a faster response time. | The segmentation algorithm is not defined. | 97.78% |
[22] | Threshold-based Adaptive Snake (AS) approach | ABCD rule + Epiluminescence microscopy (ELM) criteria algorithm | Features extracted in ABCD rule with ELM criteria. | GA + SVM with Radial Basis Function (RBF) | GA lessens the element measurements to improve framework execution. | The segmentation method is not working properly which decreases system accuracy. | 88% |
[23] | - | - | - | Multi-level CNN + BVLC Alexnet model | Easy to understand and low complexity system. | Poor efficiency of the system with a lower recall and precision rate. The segmentation and feature extraction method is not defined. | 70% |
[29] | - | Canny edge detection | The CED method detects sharp edges with image boundaries. | GA + BPNN | Cloud computing-based skin disease diagnosis system deals with enormous amounts of datasets. | Efficient segmentation is absent and system accuracy is also undefined. | - |
[36] | - | Pretrained convolutional neural network (AlexNet) | Max pooling criteria evolve with 5-layer convolutional architecture. | Deep convolution neural network + ECOC (Error-correcting output codes) linear SVM | The smart master system enhances existing work with an expansion inexactness of 3.21% in the confusion matrix. | Poor system performance due to proper segmentation approach is undefined. | 86.21% |