Pediatric sepsis is a life-threatening condition characterized by dysregulated immune responses, often leading to high mortality rates. Identifying differentially expressed genes (DEGs) is essential for understanding its pathophysiology and discovering reliable biomarkers for early diagnosis and treatment. In this study, we propose an integrative data analysis method that combines machine learning algorithms with convex hull to detect DEGs from high-dimensional gene expression datasets of pediatric sepsis patients. The convex hull approach is applied to enhance feature selection by geometrically separating relevant gene expression patterns, while supervised learning models are used to classify and validate the identified gene sets. Utilizing gene expression data from 249 pediatric patients, encompassing 11,574 genes, we propose a 10-gene signature capable of predicting sepsis-related mortality with an accuracy of 81%. Comparative experiments against baseline methods, including Principal Component Analysis (PCA) and Random Forest Feature Importance (RFFI), demonstrate that the proposed method achieves superior predictive performance with a 2–5% improvement in accuracy. This proposed method offers a promising tool for biomarker discovery and advances data-driven research in pediatric sepsis.