Investigating the impact of key tumor markers to predict and reduction of lymphoma cancer diagnosis duration with a data mining approach

Abstract

Author(s): Mohsen Ghorbian*

Various data mining techniques are available today, resulting in different results with varying precisions; therefore, selecting the appropriate methodology can result in a more complete and accurate data analysis. Hence, there are several ways to evaluate the effectiveness of data mining techniques. Choosing the appropriate data mining techniques depends on the type of data on which they will be implemented. When it comes to using data, data in every field has its significance. However, data plays a more significant aspect in specific fields, such as healthcare and data collection for caners. Using data mining techniques to analyse sensitive data like cancers can be challenging if the available information is incomplete, which can significantly impact the results. When working with the information of people with lymphoma cancer, the frequency of factors causing the disease and the lack of information are significant challenges. Lymphoma cancers can be classified as either Hodgkin's disease or non-Hodgkin's disease, which are common cancers. In this research, the criterion for selecting factors tumour markers is the presence of commonality between two types of lymphoma cancer. Five tumour markers, CD3, CD15, CD20, CD30, and LCA, along with the type of lymphoma cancer and the patient's gender, were selected as the variables of this research. Hence, to evaluate two data mining techniques, the Bayesian Networks (Naive Bayes), and the decision tree, we will apply the criteria of accuracy, sensitivity, f-score, and error ratio. However, to determine whether lymphoma cancer diagnosis factors have a positive impact, a 90% confidence interval and a 65% support value have been selected to take into account the highest level of accuracy when determining which factor is effective in diagnosing lymphoma cancer. Based on the implementation of techniques and evaluations, it was determined that the decision tree technique outperformed the Bayesian Networks (Naive Bayes) technique with an accuracy of 82.66%, a sensitivity of 94.98%, a harmonic mean of 85.36%, and an error ratio of 17.33%.Our research also concluded that the presence of CD3 and CD15 positive tumour markers, .also the gender of the individual, do not play a role in the diagnosis of lymphoma cancer. However, CD20 and LCA tumour markers can be effective in diagnosing non-Hodgkin's lymphoma, while CD30 tumour markers can be effective in diagnosing Hodgkin's lymphoma.

Share this article