C4.5 is a decision tree algorithm that can be used for making predictions. The stages start from forming a decision tree through splitting attributes, pruning and extracting rules or knowledge to then be used for prediction. However, one of the weaknesses of the C4.5 algorithm is the occurrence of overfitting and misclassification costs which result in low prediction performance. The development of the C4.5 algorithm has been carried out in terms of split attributes such as the imprecise info-gain ratio (Credal-C4.5) method using Imprecise Probability Theory, bossing gain ratio (C5.0) and average gain. This research applies the termination coefficient value (R2) as a method for modifying the gain ratio in selecting attributes as decision tree nodes which is then implemented to predict student graduation on time using a case study at the Universitas Islam Madura (UIM). Testing of the decision tree model rule for predicting student graduation on time at UIM shows that the performance values of accuracy, precision and recall are 70.49%, 77.14% and 72.97%. This performance is higher compared to the C4.5 algorithm without making modifications to the coefficient of determination, especially in accuracy and recall performance, while the precision is lower but the difference is below 1%. The difference in performance values was 11.48% (positive) for accuracy and 27.03% (positive) for recall. Meanwhile, precision performance has a difference of -0.13% (negative). The application of the Knowledge Model Rule for student graduation on time at SIMAT UIM shows very good results because it displays a prediction results page.