Dereje, Hailemariam (PhD)Endale, Mare2022-02-152023-11-042022-02-152023-11-042021-09http://etd.aau.edu.et/handle/123456789/30088Due to various reasons including the advancement of mobile devices and the proliferation of data-intensive applications, the demand for mobile data traffic is increasing rapidly. Mobile network providers are facing a challenge in improving the Quality of Service (QoS) and user experience due to ever growing data demand. Network optimization and expansion are continuous activities that enhance network quality as well as alleviates network capacity crunch. Nowadays, accurate prediction models are becoming increasingly important for predicting future data traffic demand. Anticipating data traffic demand enables operators to use it for optimization and upgrade, resulting in efficient resource utilization. In this research, a deep learning-based prediction model is proposed to predict future cellular data traffic demand using multivariate input features. The model is built with a hybrid Convolutional Neural Network (CNN) and Long Short Term Memory (LSTM) networks, called the CNN-LSTM model. Initially, the base stations are clustered using K-Means clustering based on temporal traffic patterns, and then a prediction model is developed per cluster level. The model is implemented with a deep learning library, Keras. The effectiveness of the CNN-LSTM model is evaluated using a dataset collected from ethio telecom LTE network and various metrics namely RMSE, MAPE and ��2 are used for performance evaluation. The research compared the model performance for univariate and multivariate input features cases. The results confirm that the CNN-LSTM multivariate features improved the RMSE and MAPE of the model by 58% and 50% respectively. The proposed model is also compared with CNN and SARIMA models and the proposed model outperforms both models in all evaluation metrics criteria.en-USMultivariate featuresLTE TechnologyDeep LearningCNNLSTM1D CNN-LSTMMobile Data TrafficThesis