Investigating the abnormalities of deep learning with customized architecture using deep learning 4J

Abstract

You In most cases, doctors and the use of image processing tools can assess the placenta—the actual site of physical birth—during pregnancy. Models built using Machine Learning produce performance metrics such as Accuracy, ROC, Precision, Recall, and F-Measure, which quantify this support. This research makes use of the tailored strategy built into the Weka tool, namely the DeepLearning4j package.Using common architectures such as LeNet, VGGnet, ResNet, and Alexnet yields results that are comparable. By iterating over the input layer's Loss functions and the output layer's errors, DeepLearning4j evaluates and optimizes the suggested architecture based on its quality. The experimental results show that out of seven distinct loss functions, the one with the abbreviation "MCXENT" (meaning "Multi-class Cross Entropy") produces the most accurate and least error-prone results. Additionally, a maximum accuracy of 95.7% is obtained, which is considered the best performance. These findings lend credence to an additional machine learning strategy that employs an interactive development tool for gynecologists and offer fresh assurance in its efficacy.

Keywords

• deeplearning4j
• alexnet
• placenta abnormalities
• multi-class cross entrophy
• softmax
• vgg
• loss function
• f-measure
• precision
• roc