Diabetes is a collection of metabolic illnesses caused by a persistently high blood sugar level. If a reliable estimation is achievable, diabetes risk factors and severity can be reduced. In diabetes datasets, consistent and effective diabetes prediction is challenging because of the limited amount of labeled data and the abundance of outliers (or missing values).Alongside, the incidence rates of diabetes are rising alarmingly every year. Consequently, an early diagnosis of diabetes would be the most crucial step for receiving proper treatment. Hence, a deep learning-based reorganization system has gained popularity regarding disease identification. In this work, we used an updated Convolution Neural Network (CNN) model, modifying different hyperparameters and layer topologies on the UCI 130 USA Hospitals diabetes dataset. Additionally, five different types of optimizer, namely adaptive moment estimation (ADAM), ADAMAX, A more sustainable deal has been made using the Root Mean Square Propagation algorithm (RMSprop), stochastic gradient descent (SGD), and Nesterov accelerated adaptive moment (NADAM). Furthermore, improved accuracy of 99.98% was received by the ADAMAX optimizer.