Research and Model Development for Gastric Cancer Risk Prediction

Abstract

Gastric cancer stands out as one of the most widespread deadly cancers which produces substantial rates of sickness and death throughout the world. The ability to predict gastric cancer risk early is vital to enhance patient recovery and survival statistics. The study integrated clinical and lifestyle data from public databases and simulated data which underwent preprocessing through missing value imputation and feature engineering steps including BMI creation, dietary score calculations and age grouping in combination with data balancing techniques including the Synthetic Minority Oversampling Technique (SMOTE). Three predictive machine learning models including Random Forest, Logistic Regression, and Extreme Gradient Boosting (XGBoost) underwent development and evaluation based on accuracy, precision, recall, F1-score, and Area Under the ROC Curve (AUC). The XGBoost model delivered superior performance based on experimental results achieving top scores in accuracy (0.8592), precision (0.8541), recall (0.8592), and F1-score (0.8555) which demonstrated its strong predictive power. The Logistic Regression model achieved the top AUC score of 0.8066 which demonstrates its superior ability to interpret probabilities. The study demonstrates how machine learning and specifically the XGBoost model can accurately forecast gastric cancer risk which helps enable timely medical interventions and supports tailored treatment strategies.