Schizophrenia (SZ) affects over 20 million people globally, with many patients diagnosed too late to receive appropriate treatment. Current diagnostic methods are time-consuming, requiring skilled psychiatrists, underscoring the need for more efficient approaches. This work explores using attention-based deep learning models to classify EEG signals, a non-invasive and cost-effective method, into healthy individuals and SZ patients. The proposed attention-GRU model incorporates convolutional neural networks (CNNs) for spatial feature extraction, gated recurrent units (GRUs) for sequence interpretation, and attention layers to highlight the most relevant inputs. Unlike previous works that require time-consuming manual feature extraction, our end-to-end model learns directly from EEG data, reducing preprocessing steps and enhancing the potential for real-time clinical application. Experimental results show a significant improvement in SZ detection, reaching a competitive 98.52% accuracy on an open-source EEG dataset, overcoming the accuracy reported in previous studies. This work highlights the potential of advanced deep learning models in improving the accuracy and efficiency of SZ diagnosis, addressing standardization challenges, and paving the way for more reliable diagnostic tools in psychiatric care. Our results indicate that, with further validation, AI-driven assessments can support early intervention and broader access to treatment for mental disorders.