Prostat kanseri tahmini için çeşitli denetimli makine öğrenimi tekniklerinin karşılaştırılması

Abstract

Prostate cancer is a kind of cancer that is seen worldwide and causes death of many people. Early diagnosis of cancer helps patients during the treatment phase. For this reason, cancer prediction is very crucial, according to the symptoms seen in the patient. One of the biggest problems in medicine is diagnosing diseases. The absence of certain definitive rules for the evaluation of symptoms of prostate cancer and the low rate of prediction of the diagnostic methods currently in effect made this study essential. It is thought that machine learning methods can be effective for the solution of the problems where there are no specific and definite rules and the factors affecting the event can be predicted. With this awareness, various solutions are developed by computer-aided systems. In this paper, we compare and discuss the performance of different supervised machine learning algorithms (i.e., k-nearest neighbor, support vector machines, random forest, logistic regression, linear regression, Naive Bayes, linear discrimination analysis, linear classification, multi-layer perceptron and deep neural network) for prostate cancer prediction. In this study, an open-access online prostate cancer data which consists of observations of 100 patients is used. The main intention is to evaluate the correctness in classifying data with respect to effectiveness and efficiency of each algorithm in terms of precision, recall, AUC, F1-Score, accuracy. The accuracy of the methods may vary according to the training and test data. In order to obtain more stable results, each algorithm was run more than ten times and their five best performances were recorded. The results show that multi-layer perceptron (MLP) can result in high prediction accuracy that is better compared to other approaches. Experimental results show that MLP gives the highest accuracy (97%) with the lowest error rate (0.03). The MLP classifier outperformed the other algorithms used in this study and is one of the best studies ever reported in the literature in terms of accuracy, AUC and F1 score performance criteria. As a result, we can say that if the computer is trained with machine learning methods based on patient information, it can be clinically useful with high accuracy in predicting cancer. In this way, an unnecessary biopsy of the patient can be prevented.

Towards Unsupervised Learning Driven Intelligence for Prediction of Prostate Cancer

“…From the literature, various machine learning methods have been applied towards the prediction of prostate cancer. These methods primarily hinge on a supervised learning architecture which is deemed a form of 'weak AI' and relies upon a form of external expert-based intervention to label the training sample set before the designated machine learning algorithm learns for the various data classes [17][18][19][20][21][22][23][24].…”

Optimizing explainability of Breast Cancer Recurrence using FuzzyGenetic

“…Erdem et al [6] also presented a solution for prostate cancer diagnosis comparing the machine learning models Naive Bayes (NB), Logistic Regression, k-Nearest Neighbor (k-NN), Support Vector Machines (SVM), Linear Regression, Random Forest (RF), Linear Discrimination Analysis (LDA), Multi-Layer Perceptron (MLP) and Deep Neural Network (DNN). The best result was with the MLP model, which achieved an accuracy of 97.00%.…”

Implementation of Machine Learning Mechanism for Recognising Prostate Cancer through Photoacoustic Signal

“…The effectiveness of various supervised ML techniques, including k-nearest neighbour, random forest, LR, linear discrimination analysis, linear regression, multi-layer perceptron, SVM, Naive Bayes, deep neural network, and linear classification, for predicting prostate cancer has been contrasted and addressed by Erdem and Bozkurt [ 18 ]. This study makes use of 100 patient data from an open-access Internet database on prostate cancer.…”