Patients who have been diagnosed with OSA are mainly treated with CPAP treatment. Although CPAP is considered an effective therapy for OSA patients, it is poorly tolerated and often not accepted in a large OSA patient group, limiting its overall effectiveness [1]. Unfortunately, there is an evident lack of clinical screening tools to support the early prediction of adherence to treatment in patients with OSA and help medical staff in decision-making concerning the real needs of the patients. Therefore, this study presents a methodology to identify potential features associated with the adherence to CPAP treatment using features selection methods and the development and evaluation of different machine learning models to predict CPAP therapy adherent patients. We identified different patterns of 24-hours diastolic blood pressure, daytime systolic pressure, arousal index, uric acid test and triglycerides as the relevant factors correlated with the adherence to CPAP treatment. We developed and evaluated machine learning-based models to predict CPAP treatment adherence. The best-performing model was a artificial neural network (ANN) with AUC 0.84 (95%CI: 0.72 - 0.87). The identification of potential predictors and validation of machine learning-based approaches may be helpful to identify subjects poorly adherent to therapy.