Network traffic modelling and forecasting play a crucial role in maintaining network performance, optimizing resource allocation, and providing seamless user experience. Perhaps the most widely used method for network traffic forecasting is Compound Annual Growth Rate (CAGR). While CAGR can provide a view of network traffic growth, it does not provide the required context for informed decision making that can differentiate services from the competition. Adhoc analysis of network traffic models or conventional approaches may not capture the complex network behavior that is observed today. Machine learning (ML) network traffic models can be a powerful aid to align network behavior with organizational goals. CAGR has limitations such as period choice sensitivity, network traffic volatility and seasonality. Incorrect traffic predictions increase the chance of suboptimal investments in capacity, reliability, and/or security. Network burst patterns based on high-profile events such as smartphone updates, major video game releases, or live global sporting events (e.g., the 2022 FIFA World Cup) may be overlooked. CAGR accuracy can be improved by calculating each network segment. In addition, ML network models can provide the additional context required to make informed decisions. This paper proposes using ML to enable automated iterative calculations and model attributes such as trends and seasonality, failure events, subsequent interactions between the primary and failover links, and network burst patterns. This provides the additional context that is missing from CAGR alone to make the most informed business decisions. This study considers a portion of a real Internet backbone. It analyzes traffic patterns within four consecutive years, using insights and findings to predict monthly network traffic in the fifth year. Section 2 describes the network under consideration and the data collection process. It discusses challenges incurred in this exercise and highlights influences that deteriorate data quality and the performance of any prediction. The section ends with a traffic modeling exercise and the presentation of accuracy metrics for future model evaluation. Section 3 revisits the traditional CAGR-based approach to network traffic forecasting. It exposes the limitations of the global CAGR strategy and discusses possible improvements for that method. Section 4 explores ML alternatives to network traffic forecasting. It shows how ML models can help to uncover seasonal traffic patterns and provide better forecasting. It emphasizes the importance of choosing the forecasting model appropriate to the data's nature, the presence or absence of seasonality, the prediction horizon, and the complexity of patterns in the traffic. This section also shows models in action and compares the performance of a few ML time series forecasting models when predicting traffic for the reference Internet long-haul. Section 5 summarizes critical intakes for effective and reliable forecasting and indicates future investigations. The paper ends with a list of abbreviations and references.