Predicting Cancer Treatment Response Using Machine Learning and Tabular Deep Learning

Abstract

Cancer is one of the most difficult and prevalent diseases around the globe. It is characterized by different signs among patients and their mode of treatments also varies. Despite the development of numerous advanced therapeutic approaches, selecting the most suitable treatment for each of the patients still remains one of the main problems. This study aims to develop a predictive framework according to machine learning as well as deep learning algorithms to determine a patient’s response to the cancer treatment using the Cancer Treatment Performance Dashboard Dataset.  The proposed approach involved data preprocessing, feature selection and  model training using: Logistic Regression (LR), Extra Trees (ET), CatBoost (CB), and TabNet (TN).  The framework was evaluated using 5-fold cross validation and multiple evaluation metrics including accuracy, precision, recall,F1-score and AUC-ROC. The experimental results indicated that TabNet achieved the highest performance in term of F1-score (0.90) and AUC-ROC, followed closely to CatBoost (F1-score=0.89) which reflects their ability to comply with complex nonlinear relationship in clinical data. In contrast, Extra trees demonstrated competitive performance (F1-score=0.86), while logistic regression demonstrated lower performance (F1-score=0.83) due its linear nature. This study suggests that predict patients’ treatment response at an early stage using machine learning and deep learning algorithms in analyzing clinical and therapeutic data will assist to predicting cancer treatment response to cancer and can support clinical decision-making by enabling the early identification of patients likely to benefit from specific treatments, thereby improving treatment outcomes and reducing exposure to ineffective treatments.