Prediction of Breast Cancer using Random Forest, Support Vector Machines and Naïve Bayes

Abstract

Machine learning techniques can be used to judge important predictor variables in medical datasets. This paper applies three machine learning techniques: Naïve Bayes, SVM and Random Forest to Wisconsin Breast Cancer Database. The three developed models predict whether the patients' trauma are benign or malignant. The paper aims at comparing the performance of these three algorithms through accuracy, precision, recall and f-measure. Results show that Random Forest yields the best accuracy of 99.42%, which is slightly better than both SVM and Naïve Bayes that have accuracies of 98.8% and 98.24% respectively. These results are very competitive and can be used for diagnosis, prognosis, and treatment.

A Five-Year (2015 to 2019) Analysis of Studies Focused on Breast Cancer Prediction Using Machine Learning: A Systematic Review and Bibliometric Analysis

“…The top three studies with the highest model accuracy for the BCCD were: Hernández-Julio et al 34 (accuracy = 95.90%), Singh 26 (accuracy = 92.11%) and Polat and Senturk 27 (accuracy = 91.37%). The following studies achieved highest accuracy for the WBCD: Abdar and Makarenkov 43 (accuracy = 100%), Elgedawy 41 (accuracy = 99.42 %) and Hernández-Julio et al 34 (accuracy = 99.40%).…”

“…Therefore, six papers were included in this study: three for 2016 and another three for 2017. [36][37][38][39][40][41] The 2018 search produced 5650 papers. Three sample papers were briefly read to verify selection.…”

“…Although this model appears to have obtained excellent results, this may be unreliable since no CV approach was used to handle any bias in this small dataset. Elgedawy's 41 Random Forest (RF) model ranked second in accuracy (99.42%) for the WBCD. The data allocation was 75% for training and 25% for testing.…”

“…This model used a subset of the WBCD features (Cell Size, Cell Shape, Clump Thickness and Bare Nuclei) which was first identified by the authors. As with Abdar's and Makarenkov's 43 study, Elgedawy 41 should have used some form of CV to handle the bias in the dataset. In contrast to Elgedawy, 41 Hernández-Julio et al 34 However, similar to the findings in objective 2, none of these studies seemed to have any clinical validations.…”

A Machine Learning Approach for Early Prediction of Breast Cancer

“…It is measured with dichotomous variables (in which there are only two possible outcomes).  Random Forest: Random Forest [7] is a machine learning Algorithm also called bagging algorithm based on decision trees. It is an ensemble classifier that consists of many decision trees and outputs the class that is the mode of the class's output by individual trees.…”