Comparative Analysis of Applying Imputation and Hyperparameter Optimization in Cancer Diagnosis

Abstract Cancer is one of the leading causes for death worldwide, accurate and timely detection of cancer can save lives. With more machine learning algorithms and approaches have been applied in cancer diagnosis, there is a need to analyze their performance. This study has compared the detection accuracy and speed of twenty-one machine learning and deep learning algorithms on a cervical cancer dataset. To ensure the system can potentially be applied to other types of cancers, this study has taken all features into consideration and has examined the impact of imputation and hyperparameter optimization on these algorithms’performance. The results suggest that when both imputation and hyperparameter optimization are applied, over 70% of the algorithms have improved their accuracy, among which support vector machine performs the best, with the trade-off that the execution time for most algorithmshave been lengthened. Such performance is better than when imputation or hyperparameter optimization is applied individually, and when both are absent. By not including feature selection process that most systemswould employ, this study shows such system can achieve satisfactory performance while potentially can be used to detect various types of cancers.