EGE12TH INTERNATIONAL CONFERENCE ON APPLIED SCIENCES
COMPARATIVE ANALYSIS OF MULTILAYER PERCEPTRON AND RADIAL BASIS FUNCTION NEURAL NETWORKS IN BREAST CANCER DIAGNOSIS
Yayıncı:
Academy Global Publishing House
The success Breast cancer is the most common cancer type in women worldwide, with approximately 2.3 million new cases diagnosed each year. Early diagnosis is critical for treatment success and survival rates. While the 5-year survival rate is over 90% in cases diagnosed at an early stage, this rate decreases in advanced stages. Artificial intelligence and machine learning techniques have shown promising results in breast cancer diagnosis in recent years. In this study, the performances of multilayer perceptrons (MLP) and radial basis function neural networks (RBFNN) in breast cancer diagnosis were compared. A total of 213 patients were used in the study. 56.3% (n=120) of the patients were benign and 43.7% (n=93) were malignant. The dataset was divided into training and testing data at a ratio of 70:30, and MLP and RBFNN models were applied for classification. The performance of the models was evaluated with accuracy, sensitivity, selectivity, positive and negative predictive values, F1- score and AUC. MLP model provided 97.5% accuracy, 94.4% sensitivity, 100.0% selectivity, 100.0% positive predictive value, 95.7% negative predictive value, 97.1% F1-score and 97.5% AUC values. RBFNN model achieved 91.1% accuracy, 87.0% sensitivity, 95.5% selectivity, 95.2% positive predictive value, 87.5% negative predictive value, 90.9% F1-score and 96.0% AUC values. According to the MLP model with higher success performance, the five most important risk factors were determined as age, tumor size, invasive nodules, breast quadrant and metastasis status. According to the results of artificial neural network models, risk factors associated with breast cancer were determined with high accuracy. Especially with the variable importance values provided by the MLP model, risk factors can provide early diagnosis in individuals at risk and contribute to the prevention of poor prognosis. The high classification performance of the models has the potential to provide clinicians with a valuable support tool in the decision-making process. In future studies, it is recommended to test the models with larger data sets and compare them with other algorithms.
