Modelo de inteligencia artificial hibrido para el perfilamiento de usuarios como estrategia de detección de fraudes en redes de fibra óptica

This study develops a hybrid artificial intelligence model for fraud detection in fiber optic networks under the user classification strategy, combining various machine learning approaches to improve the accuracy in the classification as fraud, anomaly or normality. Individual models such as Random Forest, Gradient Boosting and Support Vector Machine were tested. The data being worked with is provided by a telecommunications company in Norte de Santander with approximately 10 thousand records and with the following variables: Anonymized personal data (age, geographic location, user type), consumption history and network usage patterns, transactional and financial data related to billing, incident reports and service anomalies. Preprocessing is performed, the data is cleaned by eliminating null values, duplicates and outliers; then, the variables are normalized and standardized, finally, the data is divided into training sets (70%) and validation sets (30%). The results demonstrate that hybrid approaches enable more accurate analysis of user behavior in telecommunications, improving the identification of suspicious patterns in data consumption, transactions, and anomaly reporting. Compared to previous studies, which use hybrid approaches to combat telecom fraud through network analysis and predictive models, this study confirms that the combination of multiple models improves detection and reduces errors; the proposed hybrid model optimizes fraud detection in fiber optic networks, offering a good alternative for telecommunications companies that can be combined with applications in security, risk management, and revenue protection.