Enhanced Dermatoscopic Skin Lesion Classification using Machine Learning Techniques

Malicious melanoma will create greatest impact to the people. Most of the research was done on this field to find out the benign and malignant images and also to classify the different kind of skin lesions. So in the process of dermatoscopic skin lesion classification, segmentation, feature extraction, classification of the dermatoscopic images will play an important role. In this paper, we are focusing on the classification part. We are using the MNIST HAM 10000 dataset. This paper consists of two parts. In the first step, we have analyzed that always the balanced dataset will provide better accuracy compared to the imbalanced dataset. So in order to make the dataset balanced, we have used the up sampling method called Synthetic Minority Oversampling Technique (SMOTE) which greatly improves the accuracy of most of the machine learning models. Then we have analyzed the accuracy of different machine learning algorithms we have implemented. As a result, we have concluded that Support Vector Machine algorithm with Polynomial kernel provides better accuracy compared to other machine learning algorithms like Decision Tree using Gini index and Entropy, Naïve Bayes, XGBoost, Random Forest, Support Vector Machine and Logistic Regression algorithms. We have chosen Precision, Recall, F1-Score as evaluation metrics along with Accuracy. In our system, as the highest we got 96.825% accuracy with Support Vector Machine using Polynomial Kernel. Since XGBoost is a Gradient Boosting algorithm, the result provided by the algorithm might vary. In order to validate the accuracy obtained from the XGBoost algorithm, we have used the k-fold cross validation method in the XGBoost method. In our system, we have used k=10 in k-fold cross validation algorithm and we got an accuracy of 95.984% and we have concluded that SVM with Polynomial kernel provides better accuracy compared to all other algorithms.