Cellular networks face challenges in terms of real-time traffic forecasting; however, recurrent neural networks (RNNs) can be used for this effect. An RNN that has been trained rightfully may model any dynamic system; yet, learning longstanding dependencies remains the main challenge with RNN training. In this article, we present a comparative analysis of deep learning forecasting methods with focus on the two-protagonist RNN version, long short-term memory (LSTM), and gated recurrent units (GRU). We construct neural network patterns for GRU and LSTM, assess how well each model predicts the future quality of service (QoS), discuss, and compare them. These models were trained using a cellular network dataset. Based on the scenario and relevant experimental metrics results, loss, processing time, and number of parameters, there isn't a single top-notch RNN model suitable for all tasks and types of data.