Non-small cell lung cancer is a commonly diagnosed type of lung cancer. In the early stage, it is treated through surgery, but the recurrence rate is high, so it is necessary to predict the prognosis after surgery to improve the prognosis after surgery. First, clinical features are acquired from clinical data, and radiomic features are extracted from intratumoral region, peritumoral region and combined region respectively. Second, the significant clinical features and significant radiomic features are identified respectively. Third, the significant clinical and radiomic feature are concatenated and then feature selection is performed. Forth, machine-learning based classifier classify patients into recurrence and non-recurrence group. Finally, the probability of 2-year recurrence-free survival prediction is estimated by the Kaplan-Meier analysis. In experimental results, the classifier using clinical features and radiomic features showed improved performance than using features individually. Among the clinico-radiomic feature, the clinico-peritumoral 12mm features showed the highest performance, and when using the ensemble classifier with SVM and random forest, the clinico-peritumoral 12mm classification performance improved in accuracy, specificity, PPV and AUC, as 0.06%p, 7.72%p, 1.82%p, and 0.02, respectively. In the results of performance analysis according to the tumor size, the clinico-peritumoral 12mm feature showed significance in tumors under the 5cm and in case of the tumor upper 5cm, the intratumoral radiomic features showed higher performance than peritumoral radiomic features.