Esophageal cancer is a prevalent and lethal malignancy, particularly in Asian countries. Esophagotracheal fistula (ETF) is a severe complication that significantly impacts patient survival and quality of life. This study aims to develop a radiomics model based on habitat and peritumoral analysis to predict ETF occurrence in esophageal cancer patients following radiotherapy. We conducted a retrospective study involving 120 esophageal cancer patients treated between January 2018 and December 2022. We utilized computed tomography imaging data to perform habitat and peritumoral radiomics analyses. The study cohort was divided into a training set (n = 84) and a validation set (n = 36). Models were constructed using machine learning algorithms, and their performance was evaluated using the area under the receiver operating characteristic curve area under the curve (AUC), calibration curves, and decision curve analysis. The habitat-based radiomics model achieved superior predictive performance with an AUC of 0.831 in the test cohort, outperforming conventional radiomics and clinical models. The integration of habitat features, peritumoral radiomics, and clinical risk factors resulted in a comprehensive model with excellent discriminatory ability and calibration. Habitat analysis revealed distinct subregions within tumors, with specific habitats correlating strongly with ETF development. Our study demonstrates the potential of habitat and peritumoral radiomics in predicting ETF in esophageal cancer patients undergoing radiotherapy. The developed Combined model, integrating advanced radiomics features and clinical factors, provides a promising tool for individualized risk stratification and treatment planning. Future research should focus on prospective validation and the incorporation of multimodal imaging to enhance predictive accuracy.