CREATING A SEMI-SUPERVISED DATA MINING MODEL FOR PREDICTING BREAST CANCER RELAPSE

Main Article Content

Mahboobeh Zaremoayedi
Zahra Mohammadi
Mahdi Nasiri
Alireza Atashi

Abstract

Introduction: Breast cancer is one of the most common types of cancer and malignancy in Iranian women,
that recently has been a growing increase. there is always a possibility of recurrence In persons afflicted by this
disease.In regarding to the complexity of analysis, data mining is among the best solutions that is used to detect or predict cancers.


Methods: In this retrospective study, data of 809 patients with breast cancer from center of breast cancer research of Tehran’s Academic Center for Education, Culture and Research (ACECR) and 26 features from each patients were used. In regarding to high number of missing data in this collection, only information of 655 patients and 14 features of each patient were usable. Many features in records have null values, thus as one of data pre-processing and preparing phases, via Auto-Clustering algorithm, the data was divided into 10 clusters and according to dominant values in each cluster for each feature, these null features has been valuing. Data was divided to recurrence and non- recurrence classes. Semi-supervised method has been used in this study. the modeling was performed using labeled data and then a hybrid model for giving label to nonlabled data has been created. For this, data with recurrence lable divided proportional 30 percent for testing and 70 percent for training,and got to decision tree algorithms C5.0, Chaid, Quest, CRT, Autoclassifier as inputs. Then, the model was formed by mixing classifier algorithms with Confidence-Weighting-Voting method for predicting cancer recurrence, and used K-Fold (K=10) method for evaluating created model.


Results: The sensitivity of developed hybrid model was 82.93% and its specificity was 93.93%. The precision value of the model is 89.47% and its accuracy is 89.72%. this model mistakenly labelled only 10% of recurrence in patients of breast cancer as non-recurrence.


Conclusion: Creating predictive models with an appropriate sensitivity and specificity is
important since, if the possibility of recurrence is high, could perform special preventive proceedings before its spreading. The false negative percentage is also important in medical prediction models, as it can have very dangerous consequences. In the prediction model presented in this study, the value of this parameter was 10%, and in this regard, this model can be considered acceptable

Downloads

Download data is not yet available.

Article Details

How to Cite
CREATING A SEMI-SUPERVISED DATA MINING MODEL FOR PREDICTING BREAST CANCER RELAPSE. (2017). Medical Technologies Journal, 1(4), 122-123. https://doi.org/10.26415/2572-004X-vol1iss4p122-123
Section
Conference abstracts

How to Cite

CREATING A SEMI-SUPERVISED DATA MINING MODEL FOR PREDICTING BREAST CANCER RELAPSE. (2017). Medical Technologies Journal, 1(4), 122-123. https://doi.org/10.26415/2572-004X-vol1iss4p122-123

Share

Most read articles by the same author(s)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 > >>