Water resource management faces significant challenges due to hydrological droughts, emphasizing the critical need for accurate forecasting. This study meticulously analyzes hydrological droughts in the Topľa River, Slovakia, employing an innovative deep-learning model based on the Streamflow Drought Index (SDI) spanning from 1989 to 2020. The assessment of hydrological droughts involves a comprehensive data preparation phase, transforming and aligning daily discharge, water level, and temperature data with SDI requirements. This process categorizes hydrological years into dry or non-dry, laying the foundation for subsequent modeling. At the core of this research is the development of a sophisticated deep-learning model that harnesses the predictive power of SDI for precise forecasting. To overcome the challenge of limited training data, the study employs the Synthetic Minority Over-sampling Technique (SMOTE) for data augmentation, ensuring a balanced representation of both dry and normal hydrological years. The model architecture is carefully designed to predict SDI for an entire hydrological year, drawing on input features from the initial six months, which include water level, discharge, and temperature data. Implemented through TensorFlow and Keras libraries, the model incorporates strategic measures to prevent overfitting, enhancing its robustness. Through extensive training and evaluation, the model exhibits exceptional performance, achieving a remarkable 100% accuracy on both training and validation datasets. Impressively, this high level of accuracy extends to the testing dataset spanning 2010–2020, underscoring the model’s ability to generalize effectively to unseen data. This study significantly contributes to the advancement of hydrological drought prediction through the integration of deep learning and innovative data augmentation techniques. The model’s exceptional accuracy and generalizability underscores its potential as a valuable tool for drought assessment and water resource management. The research highlights the importance of an accurate data transformation process, ensuring compatibility between diverse data types and the chosen model architecture. The study’s results not only validate the model’s robust performance but also signal a promising avenue for further exploration in the realm of hydrological variability. The potential applications across diverse hydrological contexts and geographical regions emphasize the far-reaching implications of this research in contributing to our understanding of and ability to mitigate the impacts of hydrological extremes on society and ecosystems.