A Weighted Loss-Based Neural Network Approach for Imbalanced Polycystic Ovary Syndrome (PCOS) Classification

Polycystic Ovary Syndrome (PCOS) is a complex endocrine and metabolic disorder that has a major impact on fertility in women of childbearing age. PCOS is multifactorial and is characterized by various clinical manifestations, such as ovulation disorders, high androgen hormone levels, and the presence of polycystic ovaries. One of the main challenges in developing a PCOS classification model is data imbalance (class imbalance), where the number of Non-PCOS patient data is much greater than that of PCOS patients. This condition causes the model to tend to be biased towards the majority class, so that recall of the minority class is low and many cases of PCOS are not detected. This study proposes the application of Weighted Loss Function to the Neural Network model to increase the sensitivity of the model in detecting PCOS. The dataset used comes from Kaggle, consisting of 1,000 patient data with five main features. The research process includes the preprocessing stage, model design, application of Weighted Loss Function, and testing with variations in the configuration of the number of layers, epochs, and learning rates. The test results show that the combination of Weighted Loss Function with manual weighting of 1:2, learning rate of 0.001, five layers, and 50 epochs produces the best performance, with precision, recall, F1-score, and accuracy reaching 99%. These findings indicate that the proposed method is effective in improving the accuracy and sensitivity of the model to the minority class.