Year: 2025
Type: article
Abstract: In this study, a predictive model for colorectal cancer was constructed using the random forest method to improve early diagnosis and treatment. The study was based on 89,945 sample data provided by Kaggle, covering patients’ basic information, medical history, diagnosis, lifestyle and other multi-dimensional characteristics. Through data preprocessing and feature engineering, key predictors such as BMI, age, and relapse time were identified. The model significantly improved the prediction accuracy of death samples by parameter tuning and class imbalance treatment (e.g. SMOTE oversampling). The experimental results show that the random forest model performs well in the prediction of survival samples with an accuracy of 75.16%, but there is still room for improvement in the prediction of death samples. Compared to other models such as logistic regression and XGBoost, random Forest performs better on key metrics such as accuracy, recall, and F1 scores, but its model complexity and computational resource requirements are high. Through cross-validation, the study confirmed the influence of characteristics such as BMI, age, and recurrence time on the stability of model predictions. The clinical significance of the model is also discussed, indicating that it can assist physicians in optimizing treatment plans and personalized treatment decisions. Finally, the study suggests the use of larger data sets in the future to improve the generalization ability of the model.
Author: Yutong Chen
Institution: Beijing Normal University
Language: English
Subfield: Health Information Management
Field: Health Professions
Domain: Health Sciences
Open Access status: hybrid
OpenAlex ID: W4409777898