Analysis of Machine Learning Models for Breast Cancer Survival

This report analyzes the application of machine learning models in breast cancer survival prediction, focusing on the work of Montazeri et al. (2016). The study investigated the effectiveness of various machine-learning techniques, including Naive Bayes, 1-Nearest Neighbor, Trees Random Forest, Support Vector Machine, Multilayer Perceptron, RBF Network, and AdaBoost, for predicting breast cancer survival. The researchers utilized a dataset of 900 patients and evaluated the performance of these techniques based on precision, accuracy, specificity, sensitivity, and area under the ROC curve. The results showed that the Trees Random Forest model outperformed other models, achieving high accuracy, sensitivity, and area under the ROC curve. The report highlights the advantages of the Trees Random Forest model, such as its fast application, feature selection capabilities, and ease of handling missing values, concluding that it is a highly effective rule-based classification model for breast cancer survival prediction and clinical decision-making.

Running head: BIOINFORMATICS
Bioinformatics
Name of the Student
Name of the University
Author note

1BIOINFORMATICS
Breast cancer is the second most common cancer death cause among the women. The
early diagnosis and detection of breast cancer until 5 years the survival rates increases from 56%
to more than 86%. Therefore, correct tumour diagnosis is important for the treatment of breast
cancer. However, physicians face complex issues during the diagnosis procedure and so, modern
medical data diagnostic methods are used for the clinical observation. Thus, reliable and accurate
diagnosis system is necessary for the early detection and diagnosis of breast cancer (benign and
malignant). Hence, models are being designed for the diagnosis and differentiation of benign and
malignant tumours.
The paper by Montazeri et al., (2016) proposed a model called machine learning model
based on rule-based classification method for the different types of breast cancer survival
prediction. For the study, the method used was a dataset of eight attributes that included the
record of 900 patients having 876 females and 24 males and data collected from Cancer Registry
Organization of Kerman Province, in Iran. In this study, seven machine learning techniques like
Naive Bayes (NB), 1-Nearest Neighbor (1NN), Trees Random Forest (TRF), Support Vector
Machine (SVM), Multilayer Perceptron (MLP) and RBF Network (RBFN) and AdaBoost (AD)
with 10-cross fold technique was used for the breast cancer survival prediction. The performance
of these machine-learning techniques was evaluated with great precision, accuracy, specificity,
sensitivity and area under the ROC curve. The three techniques called Bagging, Random and
Boosting subspace were used for the detection of heart valve disorders and adaptive and rapid
diagnostic systems based on Learning Vector Quantization (LVQ) artificial neural network were
presented. A reinforcement mechanism was used for increasing the success rate of diagnostic
methods and reduction in decision time. For the extraction of useful information and detection of
tumors, support vector machine (K-SVM) algorithm and tools combination was presented. K-

2BIOINFORMATICS
means algorithm was separately used for the identification of hidden patterns of malignant and
benign tumors.
The findings of the study conducted by Montazeri et al., (2016) illustrated that Trees
Random Forest (TRF) technique showed impressive results as compared to other techniques
RBFN, MLP, AD, SVM, NB and 1NN. The area, sensitivity and accuracy under ROC curve of
TRF are 93%, 96% and 96% respectively. INN machine learning model gave poor performance
with sensitivity 91%, accuracy 91% and 78% area under ROC curve. The results of the study
showed that TRF model gave the best performance results and can be proposed for the survival
of breast cancer. This method has fast application and can be used for finding the effective risk
factors or feature selection alone. There is no requirement of preprocessing of data and resistance
over training in TRF. There is no requirement of rescaling, transformation or modification of the
data in TRF with natural and easy handling of missing values, computational scalability and
ability to handle irrelevant inputs.
From the above results, it can be concluded that Trees Random Forest model (TRF) is the
best model that is rule-based classification model with highest accuracy level. This model can be
recommended for breast cancer survival prediction and in effective clinical decision-making.

3BIOINFORMATICS
Reference
Montazeri, M., Montazeri, M., Montazeri, M., & Beigzadeh, A. (2016). Machine learning
models in breast cancer survival prediction. Technology and Health Care, 24(1), 31-42.