In modern telecom infrastructures, core networks are increasingly challenged by dynamic and unpredictable traffic patterns that can lead to congestion, degraded quality of service, and inefficient resource utilization. To address these issues, we propose an AI-driven approach leveraging Long Short-Term Memory (LSTM) neural networks for proactive traffic prediction and congestion prevention. Two forecasting models are developed and evaluated: one predicts traffic one hour ahead using the previous 24 h of data, while the other forecasts the next 24 h based on a seven-day history. Both models are trained on real-world core network interface data, with performance assessed using MAPE, and RMSPE metrics. Experimental results demonstrate that LSTM models achieve high predictive accuracy and consistently outperform the statistical SARIMA method across all metrics. These findings highlight the potential of LSTM-based forecasting to support intelligent traffic management strategies, enabling telecom operators to anticipate congestion and optimize resource allocation proactively.