SEQUENTIAL SELECTION OF SIGNS FOR THE CLASSIFICATION OF TEMPERATURE ANOMALIES OF BREAST CANCER BASED ON RADIOTHERMOMETRY DATA
Authors:
1.
Volgograd State University
2.
Volgograd State University
Type:
Article
Pages:
from 107
to 112
Status:
Published
Received:
31.01.2026
Accepted:
31.01.2026
Published:
31.01.2026
Language:
Russian
Keywords:
MACHINE LEARNING, LOGISTIC REGRESSION, ARTIFICIAL INTELLIGENCE, RADIOTHERMOMETRY, CLASSIFICATION
Abstract:
The application of machine learning methods in the diagnosis of breast cancer is investigated. The analysis of the formation of the feature space for the classification of temperature anomalies caused by breast cancer is carried out. The method of feature space formation based on the Sequential Floating Forward Selection (SFFB) feature addition method was used for the analysis. The results of breast examinations performed by microwave radiothermometry were used as a data set. For an objective assessment of the generalizing ability of the models, the sample is divided into training and test parts. The final verification of the results was carried out on the test data that was not involved in the learning and selection process. The effectiveness of the SFFB method was analyzed using various variations of the selection criteria, which allowed a comprehensive assessment of its flexibility. Optimization was performed using the F1 indicator, as well as a combined metric that aggregates key indicators for the artificial intelligence system: accuracy and completeness. The use of the SFFB algorithm made it possible to reduce the dimension of the feature space without significant damage to the classification quality. The initial set of 70 features has been reduced to 14 of the most informative and statistically significant variables. This result confirms the effectiveness of the SFFB method in reducing dimensionality and its ability to eliminate redundant features. The results obtained demonstrate that even with a significant reduction in dimension, it is possible to maintain almost the initial level of classification accuracy, providing faster calculations. In addition, models with fewer features have much better interpretability, which is a critical factor for making informed clinical decisions in the field of medical diagnostics, where understanding the logic of classification is often no less valuable than the result.
The application of machine learning methods in the diagnosis of breast cancer is investigated. The analysis of the formation of the feature space for the classification of temperature anomalies caused by breast cancer is carried out. The method of feature space formation based on the Sequential Floating Forward Selection (SFFB) feature addition method was used for the analysis. The results of breast examinations performed by microwave radiothermometry were used as a data set. For an objective assessment of the generalizing ability of the models, the sample is divided into training and test parts. The final verification of the results was carried out on the test data that was not involved in the learning and selection process. The effectiveness of the SFFB method was analyzed using various variations of the selection criteria, which allowed a comprehensive assessment of its flexibility. Optimization was performed using the F1 indicator, as well as a combined metric that aggregates key indicators for the artificial intelligence system: accuracy and completeness. The use of the SFFB algorithm made it possible to reduce the dimension of the feature space without significant damage to the classification quality. The initial set of 70 features has been reduced to 14 of the most informative and statistically significant variables. This result confirms the effectiveness of the SFFB method in reducing dimensionality and its ability to eliminate redundant features. The results obtained demonstrate that even with a significant reduction in dimension, it is possible to maintain almost the initial level of classification accuracy, providing faster calculations. In addition, models with fewer features have much better interpretability, which is a critical factor for making informed clinical decisions in the field of medical diagnostics, where understanding the logic of classification is often no less valuable than the result.
Keywords:
MACHINE LEARNING, LOGISTIC REGRESSION, ARTIFICIAL INTELLIGENCE, RADIOTHERMOMETRY, CLASSIFICATION
MACHINE LEARNING, LOGISTIC REGRESSION, ARTIFICIAL INTELLIGENCE, RADIOTHERMOMETRY, CLASSIFICATION
