A precise neural network approach using residual network and minority oversampling method to diagnose the correlation between colorectal cancer and food group

The key contributions of the proposed work are listed below: • Introducing the ResNet18 (Conv1D) model for CRC classification. • Performing a comparative analysis of ML and DL models, evaluating the impact of SMOTE, Borderline-SMOTE, and Adaptive Synthetic Oversampling on data balancing for minority classes. • The present research examines different food groups, including, ten subtypes under dairy products, four under meat and red meat, and two from seafood, to reveal the potential impact of consuming these foods on CRC development. • Unlike prior research, this study intended to provide recommended consumption levels for specific foodstuffs . • This insight will ensure that individuals can choose those foods and eat them in safe amounts rather than avoiding them entirely from the regular diet. In recent decades, cancer cases among individuals under 50 have increased by 79 % globally. In 2020, colorectal cancer caused one million deaths, constituting 10 % of all cancer-related mortalities. Though dietary choices influence CRC risk, the influence of certain food groups remains inconsistent. This study analyzed 76,679 males and 78,213 females aged over 59. During follow-up, 1378 men and 981 women were diagnosed with CRC. Using over 25 distinct food items in 7 major food groups as input features, five classification algorithms were applied: Extreme Gradient Boosting (XGBoost), Random Forest (RF), K-Nearest Neighbors (KNN), Recurrent Neural Networks (RNN), and Resnet18 (convolutional 1D) with ADAM optimizer. To address class imbalance in the dataset, oversampling methods such as Synthetic Minority Over-sampling Technique (SMOTE), Borderline-SMOTE (BLSMOTE), and Adaptive Synthetic Oversampling (ADASYN) were used. Our findings showed that the proposed ResNet18 with SMOTE ensemble model achieved the highest performance with 92 % accuracy, 89 % specificity, 80 % Matthews Correlation Coefficient (MCC), and a 90 % G-Mean of 90 %. The research findings revealed that higher intake of low-fat butter (RR 2.1; CI 0.93–5.23) and unprocessed red meat (RR 1.09; CI 0.87–1.38) was associated with increased CRC risk, while white meat (RR 0.77; CI 0.60–0.99) and regular butter (RR 0.97; CI 0.68–1.37) showed protective effects. Meanwhile, a dietary pattern that is rich in fish, white meat, and dairy products, along with limited portions of red meat, unprocessed meats, and soy milk, was correlated with reduced CRC risk.