IT helpdesk systems frequently face challenges in managing large volumes of support tickets, leading to classification errors, delays, and decreased customer satisfaction. While traditional machine learning algorithms have addressed some of these issues, they tend to underperform when handling complex textual data and overfit. This study proposes an enhanced ticket classification framework integrating Natural Language Processing (NLP) techniques, lemmatisation, Part of Speech (POS) tagging, Term Frequency-Inverse Document Frequency (TF-IDF) feature extraction, and dimensionality reduction via truncated Singular Value Decomposition (SVD) with deep learning models. Two recurrent neural networks (RNNs), Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), were evaluated using ticket descriptions, issues, and sub-issues. Model performance was assessed using classification metrics, including Receiver Operating Characteristic Area Under the Curve (ROC AUC), accuracy, precision, recall, and F1-score. After hyperparameter tuning with GridSearchCV, the optimised LSTM model achieved 99.20% training accuracy and 93.27% test accuracy, while the GRU-based model demonstrated faster training times with comparable performance (99.08% training accuracy, 93.02% test accuracy). Both models maintained robust generalisation, as evidenced by only marginal accuracy degradation during testing. The LSTM architecture’s memory gates effectively captured contextual dependencies, while GRU offered computational efficiency advantages for real-time applications. These results highlight that deep learning models can automate IT ticket classification with high precision, improving response times, resource allocation, and user satisfaction.