In estimating the expected wireless link channel condition for real-time mobile applications, current wireless heat maps cannot provide the required precision. In this paper we introduce a new set of LTE 'channel condition' maps that uniquely enable the radio access network to predict channel conditions of the mobile device with required accuracy. We show how to construct these channel conditions maps via machine learning without extensive driving tests. Finally we demonstrate high accuracy of wireless channel condition prediction for mobile users using our channel condition maps in conjunction with ARMAX and neural networks techniques.