Abstarct: In the era of big data and the rapid evolution of next-generation telecommunication networks, the telecommunications industry faces escalating security and financial challenges, most notably telecom fraud, which causes billions of dollars in losses annually for operators and consumers. The dynamic and evolving nature of fraudulent patterns has rendered traditional rule-baxsed systems ineffective, highlighting the urgent need for intelligent and adaptive machine learning–baxsed approaches. Recognizing the critical importance of Call Detail Record (CDR) data and the severe challenge of data imbalance in fraud detection, this study focuses on developing an advanced fraud detection architecture. The primary objective is to propose a hybrid model that leverages the strengths of classifiers such as Decision Tree (DT), K-Nearest Neighbors (KNN), and Logistic Regression (LR) to not only enhance detection accuracy but also significantly reduce the False Positive Rate (FPR), thereby improving the operational efficiency of telecommunication security systems. Key results demonstrate an F1-score of 84.58%, a ROC-AUC of 95.12% (representing a 2.35% improvement over DT), and an FPR of 0.0158, indicating a substantial reduction compared to baxseline models. These outcomes ensure an optimal balance between Precision (85.97%) and Recall (83.24%) on imbalanced datasets and highlight the model’s economic value in mitigating operational losses