In this paper we present a data-derived model of relativistic electron flux at geosynchronous orbit. The model is driven by multiple solar wind and magnetospheric inputs and combines the deterministic approach of nonlinear dynamics with conditional probability consideration. The model is used for one-day predictions of daily flux maxima for the years 1995--2000. The deterministic part of the model yields average prediction efficiency of 0.77 and linear correlation coefficient of 0.89. It identifies solar wind velocity and SymH index as most relevant input parameters. The probabilistic part of the model quantifies the risks of deviations from deterministic predictions as a function of average solar wind and magnetosphere conditions.