Brain cancer remains one of the most aggressive and life-threatening diseases, where timely and accurate subtype classification plays a critical role in improving patient outcomes. However, gene expression (GE) -based diagnostic methods often suffer from challenges such as high dimensionality, limited sample sizes, and noisy data, which hinder their accuracy and clinical applicability. This research proposes a deep learning (DL)-based framework, DeepGEx, to enhance the classification performance of brain cancer subtypes using GE data. The framework integrates Random Forest (RF) for feature selection (FS), Autoencoders (AE) for dimensionality reduction (DR), and Generative Adversarial Networks (GAN) for data augmentation (DA). The enriched GE data is then classified using a custom-designed Deep Neural Network (DNN). The model was trained and evaluated on the publicly available GSE50161 dataset. DeepGEx achieved a training accuracy of 97%, validation accuracy of 92%, and test accuracy of 92.31%. The area under the ROC curve exceeded 0.90 for all cancer subtypes, demonstrating strong discriminative capability. The confusion matrix indicated less than 5% misclassification across classes. Feature importance analysis further revealed biologically meaningful gene markers, supporting clinical interpretability. Compared to existing methods, DeepGEx offers a more accurate, interpretable, and scalable solution, making it a promising AI-driven approach for brain cancer diagnostics.