This paper studies a novel fuzzy deep learning network (FDLN) to analyze electroencephalogram (EEG) signals in order to classify and diagnose major depressive disorder (MDD). The research was performed by designing a hybrid model that combined fuzzy DL model logic, NSGA-II (Non-Dominated Sorting Genetic Algorithm II) and EEG classification to make a framework called EEG-PDL, which collected data from patients clinically diagnosed with MDD and from healthy controls. The findings demonstrate that this new model significantly outperforms conventional deep learning and machine learning approaches. Specifically, it achieved an accuracy of 97.47%, with a sensitivity of 97.92% and a specificity of 97.03%, while other standard models such as CNNs, LSTMs, and SVMs consistently performed lower. The fuzzy component gave the model an advantage in handling the uncertainty and noise that naturally occur in EEG data, while deep learning made it possible to automatically extract subtle and complex patterns in brain activity that traditional methods would miss. The combination of this NSGA-II optimization with backpropagation helped the researchers advance interpretability and reliability for diagnosis.