This paper presents a method of International voice call fraud detection in the case of the ethiotelecom. International voice call fraud is a type of voice call which used from both user sides without the use of data. Telecom fraud is the main challenge from the start of telecom service. Many elements affect telecom voice fraud. Some of them include the adoption of new technology without first assessing the security gap, ignorance of the underlying causes of fraud, and the evolving fraud behavior. This type of call came from different operators or called different operators as regular voice calls by changing digit size, changing TG code, by hidden number, and exchanging country code or operator code. The detection of such as voice call fraud, CDRs are classified into subtasks for analysis. The effectiveness of three machine learning algorithms—Logistic Regression Classifier, Linear SVC, and Random Forest Classifier—in identifying voice calls is analyzed in this thesis. Voice call fraud separated incoming and outgoing calls based on CDR as a classification task. 741,749 CDR are captured for controlled laboratory environment tool and by different validation techniques analysis separated into normal call and Fraud call. Out of 741749 CDRs data, the success 151,169 CDRs data are used for analyses. Comparing Random Forest Classifiers to the left two methods, which have the best performance, the overall classification accuracy that they accomplish is 99.99%. Keywords: Machine learning algorithms, International voice call fraud number