Diabetes mellitus, a critical global health concern, is characterized by irregularities in insulin secretion and glucose level regulation. Maintaining balanced plasma glucose levels is essential for optimal organ functionality. Existing artificial pancreas models often fail to adapt to individual dietary variations, limiting their effectiveness. This study proposes a deep learning model (CODLSTM) that integrates CNN-1D and LSTM networks for improved performance. Unlike traditional models with fixed parameters, our approach dynamically adapts to unique dietary behaviors. A comprehensive dataset simulating diverse human behaviors is used to evaluate the proposed architecture. Experimental results demonstrate that the proposed CNN-1D + LSTM model achieves an average absolute error of 19.7, significantly outperforming standalone CNN-1D (91.2) and LSTM (53.4) models. This superior predictive accuracy underscores the efficacy of the proposed approach for personalized glucose level management, paving the way for tailored healthcare strategies in diabetes management.