Prediction of Prostate Cancer by Using Data Mining Techniques with Artificial Neural Network

Abstract

The idea is to explore the potential of using machine learning algorithms in predicting prostate cancer. The estimation models of prostate cancer should be made appropriately to increase the chance of prostate cancer survival. Prostate-cancer is a highly prevalent cancerous disease in men. Therefore, recognition of prostate-cancer as early as possible is needed and essential so that practitioners can establish probable outcomes and set effective strategies to treat the patients. The dataset used was the original database with 76,683 records of patients. The data were acquired from the National Cancer Institute (NCI). The number of patients after processing, cleaning, and eliminating duplicate data stands at 36,159. The data for the produced artificial neural network (ANN) was provided through the selection of data attributes, e.g., depending on the patient's age and the levels of PSA. We used a methodology that divided the data into 70% training and 30% test and validation sets, respectively. AUC and ROC curves are applied to evaluate the experimental outcomes of the proposed model. An ROC curve is a standard method of illustrating the performance of a binary classifier in terms of its accurate positive and false positive rates over a threshold range. The AUC provides a point of reference, giving a single number to represent the overall performance. Our in-depth ANN, educated by means of scaled conjugate convergence feed-forward backpropagation, achieved 91.50% in the overall confusion matrix. The best validation is 0.0644 at measure 26, the gradient is 0.0151 at measure 32, and the validation check is six at measure 32. ANN was close in predicting prostate cancer and did not use a biopsy. Nevertheless, we believe that ANN performance is insufficient for application in clinical practice. This paper observed that prostate cancer can be effectively predicted using machine learning methods. Nevertheless, incorporating more variables, including life factors that are suitable to estimate prostate-cancer, could also be a good alternative that would enable the ANN to perform better.