Diabetes forecasting is a rather important aspect given that through forecasts, the healthcare providers are able to predict new needs of the patient and adjust interventions. Standard techniques are not effective since they do not account for a progression, which is stochastic and thus not strictly linear, of diabetes. This research aims to develop a new forecasting model called HybridFeatureNet which is presented in this research to address the current limitations of diabetes forecasting using conventional deep learning techniques in conjunction with feature extraction methods such as NMF and PCA. These methods work hand in hand to provide hidden characteristics and feature reduction, thereby improving the dataset’s forecasting brilliance by integrating a convolutional neural network (CNN) model. The proposed approach successfully solved the problem with 87% accuracy in predicting the further development of diabetes, which benefited from the work compared to traditional approaches. These outcomes validate the utility of the proposed model in enhancing forecast performance and deriving human-interpretable pattern understandings of diabetes progression. In this study, practical use of HybridFeatureNet in clinic is presented, encouraging more the use of this type of innovation for better group and individual diabetic control. Subsequent study will focus on refining the framework and assessing its extension to other chronic illnesses with the aspiration of improving the precision of various models and the decision-making processes within the sphere of health care services delivery.