Predicting blood glucose variability accurately is essential in managing type 1 diabetes mellitus (T1DM). The occurrences of hypoglycemic and hyperglycemic events can be managed much better if detected early on, therefore preventing severe adverse events. Often existing deep learning methods either cannot capture local glucose variation patterns or do not model long-term temporal dependencies from the continuous glucose monitoring (CGM) data. To overcome this limitation, this study presents a novel multi-scale convolutional neural network-bidirectional long short-term memory-transformer (MSCNN-BiLSTM-Transformer) hybrid framework for multi-class glucose state prediction. In sequential glucose measures? the proposed model aims at learning a discriminative representation through multi-scale convolution-based feature extraction, bidirectional temporal learning and transformer based self-attention mechanism. The OhioT1DM dataset was utilized in experiments that employed 12 previous glucose readings as inputs to classify glucose states into three states, hypoglycemia, normal and hyperglycemia. The proposed model achieve an accuracy of 97.59%, precision of 97.78%, recall of 97.56%, F1 score of 97.60%, ROC-AUC of 99.79%, Matthews correlation coefficient (MCC) of 95.30%, Cohen’s Kappa 95.76%, and outperform other conventional CNN, LSTM, GRU, Transformer and other baseline models. This work presents an ablation study that showed our multi-scale convolution, bidirectional sequence learning and Transformer contextual modelling and feature fusion is effective. Moreover, the SHAP-based explainability analysis sheds light on how historical glucose data affected the final prediction, demonstrating the proposed framework's clinical reliability and interpretability. The results obtained from the tests show that the revealed hybrid model is a robust, accurate, and interpretable model for intelligent glucose state prediction which has the potential for being used as a decision-support tool for personalized diabetes treatment and other smart healthcare solutions in the future.