Sepsis is a major health issue affecting newborns, often caused by factors such as premature birth, birth asphyxia, pneumonia, and meningitis. Early identification of sepsis is critical for improving neonatal outcomes, but it varies based on birth history, vital signs, laboratory results, and other clinical features. A robust classification model is built to identify sepsis in newborns under one year of age, using data from the Children's Hospital of Philadelphia (CHOPS) and the Mendeley Neonatal Repository (MENR). Both datasets were analyzed from various perspectives, and a balanced dataset was obtained using stratified random sampling. To enhance feature interpretability, SHAP (SHapley Additive exPlanations) was used to identify the most important features influencing sepsis predictions This approach helps to clarify the contributions of individual features, making the model's predictions more understandable and transparent. Data triangulation was employed to extract the optimized features from CHOPS and MENR to create a new optimized dataset (OPTCM). A hybrid stack ensemble classifier was designed to test the performance of all the models. The optimized model (OPTCM) demonstrated superior performance with a 99.2% accuracy, outperforming the original CHOPS dataset's 95.3% and MENR's 96.6%. The improved accuracy of the optimized model suggests that it could serve as an efficient tool for classifying patterns for early sepsis detection in newborns, potentially improving clinical outcomes.