An anomaly in 3G/4G networks represents an unusual traffic pattern or behavior that significantly deviates from the normal (or usual) network behavior. However not all anomalies are known to LTE networks since existing anomaly detection systems in wireless networks employ supervised algorithms to classify and detect abnormal network behaviors. Self-Organizing Network (SON) in 3GPP 36.902 requests 3G/4G networks to regularly self-optimize parameters and algorithmic behavior in response to observed network performance issues. Thus unsupervised algorithms are desirable to automatically characterize the nature of traffic behavior and to detect traffic anomalies from normal network patterns in3G/4G LTE networks. We apply a Bayesian probabilistic algorithm to detect traffic anomalies in the generating network entities with high sensitivity and complexity. With a demand of early detection before happening, we also develop a sandwich alike algorithm to detect traffic and service anomalies in advance with a likelihood estimation. Experiments validates early detection and shows low false alarm rate in our solution along with low complexity in production in 3G/4G LTE networks.