Abstract: The precision of attitude sensors in heave compensation systems for floating operations directly affects the compensation accuracy. Mainstream attitude sensors typically achieve a heave measurement accuracy of 5 cm or 5%, which limits the system's ability to reach a 95% compensation rate. To address this issue, a method based on Support Vector Classification (SVC) was developed to correct measurement errors. The attitude sensor was mounted on a six-degree-of-freedom platform, and the platform's heave displacement, measured by a dynamic capture system, was used as the reference ground truth to determine the sensor's measurement errors. Through significance analysis of the relationship between the errors and the sensor's output signals, three parameters—acceleration, velocity and displacement—were selected as inputs for the SVC model. An SVC model was constructed in MATLAB, and data collected from the system were used for model training and parameter optimization, employing Bayesian optimization and k-fold cross-validation techniques. The trained SVC model was then integrated into data acquisition software, with custom code developed to implement classification functions.The results demonstrate that the SVC model effectively identified the error ranges corresponding to the three output parameters of the attitude sensor, enabling specific correction values for each error category. The correction process required an average time of only 1.4 ms. After correction, the root mean square error of the heave displacement decreased from 0.4567 cm to 0.2622 cm (or 2.622% of the amplitude) , while the absolute mean error was reduced from 0.3668 cm to 0.1989 cm. These findings indicate that the proposed method significantly reduces heave measurement errors. Furthermore, the method is applicable for correcting errors in other degrees of freedom outputs of the attitude sensor.