Skin cancer is one of the most strongly growing types of cancer due to pollution and many other factors. The early diagnosing of skin cancer and its types can contribute to stop this rapid growth of skin cancer disease and for this purpose AI can be utilized. Skin cancer is one of the most researched topics, with various methods used to diagnose it; however, there is always room for improvement and new technologies to perform this task more effectively. This study aimed to utilize a data set containing skin cancer types, known as PAD-UFES-20, to classify different types of skin cancers using advanced data pre-processing and a combination of deep learning and machine learning techniques and finally these features were analyzed to determine what impacted most for disease to be classified as skin cancer. The proposed methodology includes detailed pre-processing of the data set, a custom Convolutional Neural Network model for feature extraction, training Boosting models on pre-processed data, and finally finding features that impact the model the most to identify disease to be a skin cancer type. The CatBoost Classifier, XGBoost Classifier, and LGBM Classifier were trained on the PAD-UFES-20 data set to diagnose skin cancer and its six different types. With better pre-processing techniques, we obtained more accurate results compared to previous studies. The XGBoost Classifier produced the highest accuracy compared to CatBoost and LGBM Classifiers. In addition, this study also includes research on the features of the data set that most effect the prediction of the model. In summary, the proposed method focused more on the best pre-processing and feature extraction techniques to obtain the most possible predictions from Boosting models.