Breast Cancer Risk Prediction with Stochastic Gradient Boosting

Mehmet Kivrak^1*

¹Department of Biostatistics and Medical Informatics, Faculty of Medicine, Recep Tayyip Erdogan University, Rize, Turkey.

Abstract

Breast cancer, which is an important public health problem worldwide, is one of the deadliest cancers in women. This study aims to classify open-access breast cancer data and identify important risk factors with the Stochastic Gradient Boosting Method. The open-access breast cancer dataset was used to construct a classification model in the study. Stochastic Gradient Boosting was used to classify the disease. Balanced accuracy, accuracy, sensitivity, specificity, and positive/negative predictive values were evaluated for model performance. The accuracy, balanced accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and F1 score metrics obtained with the Stochastic Gradient Boosting model were 100 %, 100 %, 100 %, 100 %, 100 %, and 100 %, and 100 % respectively. In addition, the importance of the variables obtained, the most important risk factors for breast cancer were a cave. points_mean, area_worst, and perimeter_worst, concave. points_worst respectively. According to the study results, with the machine-learning model Stochastic Gradient Boosting used, patients with and without breast cancer were classified with high accuracy, and the importance of the variables related to cancer status was determined. Factors with high variable importance can be considered potential risk factors associated with cancer status and can play an essential role in disease diagnosis.

Keywords: Breast cancer, Machine learning, Ensemble learning, Stochastic gradient boosting

In this study, we aimed to develop an algorithm that classifies benign and malignant cells with high accuracy by using the stochastic gradient boosting method with digitized values of cell nuclei obtained from fine needle aspiration (FNA) images on the breast mass.

Although machine learning gives good results in the classification of histopathological images of breast cancer, the increase in data size and architectural edifice affects the category performance results of these techniques. Ensemble-learning systems, a sub-branch of machine learning, on the other hand, are layer-based learning machines than traditional machine learning layers. Therefore, the systems perform better in classifying cancer images compared to ordinary machine learning techniques generated by ensemble learning algorithms. an important feature of community learning. This significant improvement will further facilitate the widespread adoption of ensemble learning in medicine.

In a recent study, support vector machines, k-nearest neighbors, and probabilistic neural network classifiers were used to differentiate benign and malignant tumors of the breast with signal-to-noise ratio feature sequencing, sequential forward selection-based feature selection, and principal component analysis with feature extraction. are combined. The complete accuracy of breast cancer diagnosis is achieved with a support vector machine classifier model against two commonly used breast cancer comparison datasets equaling 98.80 % and 96.33 %, respectively.^[10]

In another study, two different classifiers for breast cancer classification were presented: Naive Bayes (NB) classifier and nearest neighbor (KNN). A comparison of two novel implementations was proposed and their accuracy was evaluated using cross-validation. The outcomes illustrate that KNN gives the highest accuracy (97.51 %) with the lowest error rate after the NB classifier (96.19 %).^[11]

In an article, a performance comparison between different machine learning algorithms, Support Vector Machine (SVM) in Wisconsin Breast Cancer, Decision Tree (C4.5), Naive Bayes (NB), and k Nearest Neighbor (k-NN) (original) data sets were made. The main purpose is to evaluate the accuracy of each algorithm in classifying data according to its efficiency and effectiveness in terms of accuracy, precision, sensitivity, and specificity. Experimental outcomes display that SVM provides the maximum accuracy (97.13 %) with the lowest error rate. All experiments are conducted in a simulated environment and executed with the WEKA data mining tool.^[12] Similarly, Aamir et al. developed a cancer risk classification model using supervised machine learning methods such as SVM, Random Forest, Gradient Boost, Artificial Neural Network and Multilayer Perceptron Model in the study called Predicting Breast Cancer Leveraging Supervised Machine Learning Techniques. With an accuracy of 99.12%, the Multilayer Sensor Model outperformed all other models reviewed.^[13]

In another paper, models were created using machine learning techniques to detect and visualize important prognostic indicators of breast cancer survival rate. In terms of both model accuracy and calibration measurement, all algorithms produced close results from the lowest decision tree (accuracy = 79.8) and highest random forest (accuracy = 82.7).^[14] Similarly, in their study on an ensemble machine learning algorithm to predict driver genes in metastatic breast cancer, mirsadeghi et al. examined somatic mutation data from 450 metastatic breast tumor samples from the Bio Cancer Genomics Portal. First, the drivers and passengers predicted by SVM, ANN, RF and EARN are introduced. Comparison between outcomes shows that ROC-AUC reaches 99.24% when EARN is used for MBCA and 99.79% for breast cancer.^[15]

An important difference between the proposed studies from the aforementioned studies is the open source analysis in the R program. The proposed algorithm shows that the benign and malignant breast cancer classification performances of the SGB model are high. Therefore, it is suggested that SGB architecture can be used in the classification of benign and malignant breast cancer.

Although the model has high classification performance metrics, the system can be retrained by feeding more data or applying other ensemble learning methods to increase model validity and reliability. Thus, it is aimed to obtain high-precision diagnostic results by integrating the developed model into decision-making processes.

Conclusion

Negative emotional responses to stressful life events are a normal short-lived emotional-behavioral response to threat; Anger and fear are not signs of a psychopathological disorder. Breast cancer can have a significant impact on patients' lives, including their health-related quality of life. In addition, psychological responses to diagnoses and treatments can vary significantly over time and with clinical aspects of the healthcare pathway. Throughout the diagnosis and survival process, breast cancer patients report significant declines in their psychological health and quality of life, including greater pain and fatigue, insomnia, and greater interference of stressors (physical and emotional) with social activities.^[16]

In this study, a boosting algorithm with a k-fold cross-validation resampling method using all data was used. High accuracy classification of the variables obtained by digitizing the histopathological images with the Stochastic Gradient Boosting method, which is an ensemble learning algorithm, is provided.

In summary, according to the results obtained from the study, it was determined that the model built and designed on boosting gave promising predictions in the classification of breast cancer (benign/malignant) based and could be used for this.

Acknowledgments

In this study, I would like to express my deepest gratitude to the Machine Learning Repository (UCI) open access databases that provided the datasets.

Conflict of interest

None.

Financial support

None.

Ethics statement

 Ethics committee approval is not required as a retrospective study is planned on data sets obtained from open access databases.

1.      Bray F, Ferlay J, Soerjomataram I, Siegel RL, Torre LA, Jemal A. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2018;68(6):394-424. doi:10.3322/caac.21492

2.      Gazioğlu D, Büyükaşik O, Hasdemir AO, Kargici H. BIRADS 3 ve 4 meme lezyonlarına yaklaşım: Hangi olgulara biyopsi yapılmalı?. J Turgut Ozal Med Cent. 2009;16(1):19-24.

3.      Akpınar H. Veri tabanlarında bilgi keşfi ve veri madenciliği. İstanbul Üniv İşlet Fak Derg. 2000;29(1):1-22.

4.      Polikar R. Ensemble learning. InEnsemble machine learning 2012 (pp. 1-34). Springer, Boston, MA.

5.      Wolberg WH, Street WN, Heisey DM, Mangasarian OL. Computer-derived nuclear features distinguish malignant from benign breast cytology. Hum Pathol. 1995;26(7):792-6. doi:10.1016/0046-8177(95)90229-5

6.      Arslan A, Şen B. Detection of non-coding RNA's with optimized support vector machines. In2015 23nd Signal Processing and Communications Applications Conference (SIU) 2015 May 16 (pp. 1668-1671). IEEE. doi:10.1109/SIU.2015.7130172

7.      Schapire RE. The boosting approach to machine learning: An overview. Nonlinear estimation and classification. 2003:149-71.

8.      Sun CK, Tang YX, Liu TC, Lu CJ. An Integrated Machine Learning Scheme for Predicting Mammographic Anomalies in High-Risk Individuals Using Questionnaire-Based Predictors. Int J Environ Res Public Health. 2022;19(15):9756. doi:10.3390/ijerph19159756

9.      Friedman JH. Stochastic gradient boosting. Comput Stat Data Anal. 2002;38(4):367-78.

10.   Osareh A, Shadgar B. Machine learning techniques to diagnose breast cancer. In2010 5th international symposium on health informatics and bioinformatics 2010 Apr 20 (pp. 114-120). IEEE. doi:10.1109/HIBIT.2010.5478895

11.   Amrane M, Oukid S, Gagaoua I, Ensari T. Breast cancer classification using machine learning. In2018 Electric Electronics, Computer Science, Biomedical Engineerings' Meeting (EBBT) 2018 Apr 18 (pp. 1-4). IEEE. doi:10.1109/EBBT.2018.8391453

12.   Asri H, Mousannif H, Al Moatassime H, Noel T. Using machine learning algorithms for breast cancer risk prediction and diagnosis. Procedia Comput Sci. 2016;83:1064-9. doi:10.1016/j.procs.2016.04.224

13.   Aamir S, Rahim A, Aamir Z, Abbasi SF, Khan MS, Alhaisoni M, et al. Predicting Breast Cancer Leveraging Supervised Machine Learning Techniques. Comput Math Methods Med. 2022;2022. doi:10.1155/2022/5869529

14.   Ganggayah MD, Taib NA, Har YC, Lio P, Dhillon SK. Predicting factors for survival of breast cancer patients using machine learning techniques. BMC Med Inform Decis Mak. 2019;19(1):1-7. doi:10.1186/s12911-019-0801-4

15.   Mirsadeghi L, Haji Hosseini R, Banaei-Moghaddam AM, Kavousi K. EARN: an ensemble machine learning algorithm to predict driver genes in metastatic breast cancer. BMC Med Genomics. 2021;14(1):1-9.

16.   Ranieri J, Di Giacomo D, Guerra F, Cilli E, Martelli A, Ciciarelli V, et al. Early Diagnosis of Melanoma and Breast Cancer in Women: Influence of Body Image Perception. Int J Environ Res Public Health. 2022;19(15):9264. doi:10.3390/ijerph19159264