The increase in the number of users and the high bandwidth demand of the Internet have increased network traffic at an alarming rate. Currently, social interaction among people has become highly dependent on Internet applications. In addition, government offices, schools, non-governmental organizations, and the private sector automate their services through digital applications. This increase in data demand, which is a source of revenue for Internet service providers (ISPs), must be managed properly in order for them to provide a higher level of service to their customers. One of the solutions to this issue is to conduct a predictive analysis of the network traffic to understand the traffic characteristics, explore the utilization levels of links and network elements' capacity to identify the bottleneck uplinks that need optimization to increase the network performance. Prediction and modeling research conducted on the Addis Ababa city network are based on mobile data, and the output may not entirely explain the unique characteristics of the fixed network behavior. Therefore, the aim of this research is to model and forecast the fixed access network data by exploring two well-known predictors: the recurrent neural network based Long-Short-Term Memory (LSTM) and the statistical Seasonal Auto-Regressive Integrated Moving Average (SARIMA) algorithm. In addition, a temporal traffic network characteristic pattern analysis has been conducted. Six months of hourly data from a fixed access network was collected by adding sensors to the network management system. Seasonality, non-linearity, and trends were observed in the aggregated fixed network data tests. To evaluate forecast performance accuracy as well as to compare the two models, error performance measuring metrics such as R squared, Root Mean Squared Error (RMSE), Mean Average Percentage Error (MAPE), and Mean Absolute Error (MAE) were used. The thesis's results confirmed that both algorithms are effective and yield satisfactory performance in forecasting non-linear and seasonal fixed access network traffic data. However, LSTM was more accurate than the SARIMA model when evaluated using an error performance matrix. The result of this thesis could be used as an input for planning, designing, and optimization of fixed access networks.