Cancer is one of the most common diseases worldwide. The development of early and accurate diagnosis methods for cancer is of great importance to people. Extraction of radiomic features, which involves extracting numerical features from imaging-based data, is one method that can be useful for cancer detection and treatment. In this study, classification was performed on multi-class cancer types using radiomic features that are extracted from medical images. Different machine-learning algorithms were employed in the classification process, and their hyperparameters were optimized using the Optuna library. The models were retrained using the best parameters obtained and evaluated based on performance metrics, including accuracy, precision, sensitivity, and F1-score, on the test data. In addition, SHAP (Shapley Additive Explanations), one of the explainable artificial intelligence (XAI) approaches, was employed to make the decision mechanisms of the models more understandable. The effect of the variables was visualized for each model, and the most decisive features were identified. As a result of the study, accuracy rates of 87%, 86%, and 81% were achieved with the XGBoost, LightGBM, and CatBoost algorithms, respectively. This holistic approach improves model performance in radiomics-based classification systems and strengthens the clinical validity of AI-based decisions.