Abstarct: Security and financial integrity in the telecommunications industry are of critical importance, particularly given the massive and continuously growing volume of Call Detail Record (CDR) data. Timely detection of telecommunication fraud represents a complex challenge due to the inherently imbalanced nature of the data, where fraudulent cases are significantly outnumbered by legitimate transactions. This challenge necessitates the adoption of robust and effective machine learning approaches. This study proposes a hybrid ensemble learning frxamework that integrates AdaBoost and Random Forest algorithms to develop an efficient and imbalance-resilient telecommunication fraud detection system. The proposed hybrid model leverages the strength of AdaBoost in enhancing weak learners and the capability of Random Forest in reducing variance and mitigating overfitting, thereby maximizing predictive accuracy in imbalanced data environments. To address the class imbalance issue in CDR data, data preprocessing and resampling techniques such as the Synthetic Minority Over-sampling Technique (SMOTE) are employed. The proposed model is evaluated on real-world CDR datasets. Experimental results demonstrate that the AdaBoost–Random Forest hybrid model significantly outperforms individual baxseline models (standalone Random Forest and AdaBoost). The model achieves superior performance across key evaluation metrics, including F1-score = 0.86, Recall = 0.87, Precision = 0.86, PR-AUC = 0.87, and ROC-AUC = 0.98, compared to the non-balanced setting (F1-score = 0.85 and Recall = 0.84). These results confirm the model’s effectiveness in successfully identifying over 87% of fraudulent cases, offering a robust and practical solution for combating fraud in telecommunication networks.