Increasing Quality in Deep Learning Models through Stratification

Abstract

In this study, increasing quality in deep learning models through data stratification was investigated, focusing on a dataset of patients with diabetes mellitus. An artificial neural network (ANN) was created with the objective of predicting the probability of diabetes complications based on multiple clinical and biomedical characteristics of the patients. To improve model quality, a data stratification approach was implemented, dividing the data set into homogeneous groups to ensure balanced representation of the classes of interest in the training and testing data sets. Data stratification proved to be an effective strategy to improve the quality of the deep learning model. By maintaining a balanced distribution of diabetes complication classes in the training and testing data sets, the model was able to learn more representative patterns and better generalize to unseen data. This resulted in a significant improvement in the accuracy and generalizability of the model, resulting in more accurate and useful predictions for early identification of complications in patients with diabetes mellitus. Together, these findings highlight the importance of data stratification in increasing quality in deep learning models, especially in clinical applications such as predicting medical complications. This approach has the potential to improve healthcare by enabling more accurate and earlier identification of risks for patients, which can lead to more effective interventions and better overall health outcomes.