M.S. AAI Capstone Chronicles 2024

CNN Lung Disease Classification

13

confirming the problem of overfitting that we needed to address.

In response to the overfitting observed in the initial model, dropout regularization was introduced. The inclusion of dropout regularization resulted in a slightly more stable model with minor improvements in validation accuracy. However, overfitting was still evident after several epochs, as indicated by the figure below. While training loss continued to decrease and training accuracy increased, validation loss did not show a corresponding decline, highlighting that overfitting is a persistent challenge in our analysis.

171