The aim of this study was to develop a simple, fast and efficient clinical diagnostic model, composed of exercise stress echocardiography (ESE) indicators, of the exercise capacity of patients with chronic heart failure (CHF) by comparing the effectiveness of different classifiers. Eighty patients with CHF (aged 60±11years; 78% male) were prospectively enrolled in this study. All patients underwent both cardiopulmonary exercise test (CPET) and ESE and were divided into two groups according to the VE/VCO2 slope: 30 patients with VE/VCO2 slope ventilation classification (VC)1 (i.e., VE/VCO2 slope<30) and 50 patients with VC2 (i.e., VE/VCO2 slope≥30). The analytical features of all patients in the four phases (rest, warm-up, peak and recovery phases) of ESE included the following parameters: left ventricular (LV) systolic function, LV systolic function reserve, LV diastolic function, LV diastolic function reserve and right ventricular function. Logistic regression (LR), extreme gradient boosting trees (XGBT), classification regression tree (CART) and random forest (RF) classifiers were implemented in a K-fold cross-validation model to distinguish VC1 from VC2 (LVEF in VC1 vs. VC2: 44±8% vs. 43±11%, P=0.617). Among the four models, the LR model had the largest area under the curve (AUC) (0.82; 95% confidence interval [CI]: 0.73 to 0.92). In the multiple-variable LR model, the differences between the peak-exercise-phase and resting-phase values of E (ΔE), s'peak and sex were strong independent predictors of a VE/VCO2 slope≥30 (P value: ΔE=0.002, s'peak=0.005, sex=0.020). E/e'peak, ΔLVEF, ΔLV global longitudinal strain and Δstroke volume were not predictors of VC in the multivariate LR model (P>0.05 for the above). Compared with the LR, XGBT, CART and RF models, the LR model performed best at predicting the VE/VCO2 slope category of CHF patients. A score chart was created to predict VE/VCO2 slopes≥30. ΔE, s'peak and sex are independent predictors of exercise capacity in CHF patients.