Machine learning (ML), deep learning (DL), and artificial intelligence (AI) are increasingly employed in medical assistance and statistical analysis. These technologies enhance medical diagnostics by enabling precise image segmentation, feature extraction, and lesion localization. Such capabilities significantly improve the accuracy of auxiliary diagnoses and facilitate more effective treatment planning. In the field of medical statistics, machine learning (ML), deep learning (DL), and artificial intelligence (AI) can aid in large-scale analysis and mining of medical data to identify potential disease patterns and predict patient risks. For instance, doctors use ML and AI to integrate and analyze patients’ genomic and imaging data for personalized treatment plans. However, extensive research has not been carried out concerning physical examination data in this field. However, the early stages of numerous diseases are not objectively reflected in the current physical examination evaluation system. For instance, with gastric cancer, the significance of routine physical examination detection indexes is limited, making it challenging to diagnose precisely during consultations. This factor significantly complicates the early detection of gastric cancer. In our study, we introduce a method based on PSO and Stacking Learning (PSOS) for predicting early-stage gastric cancer. This method leverages basic personal bioinformation and standard physical examination data to construct a predictive model. Designed to assist in the pre-gastroscopy clinical screening of gastric cancer, this model aims to identify potential cases before the need for invasive procedures such as gastroscopy and tissue biopsies. We trained the model using data from 22,304 actual patients and achieved a high success rate in early detection, demonstrating the model’s robust capability to identify gastric cancer at early stages.