In this paper, an extended state observer (ESO) based data-driven sliding mode predictive learning control (SMPLC) scheme is proposed for a class of nonlinear batch processes with unknown dynamics, aiming to mitigate the impact of nonrepetitive external disturbances. The batch process is firstly described by an iterative dynamic linearization data model (IDLDM) with a residual term. A P-type sliding surface is then designed to establish an iterative sliding mode control scheme, where the generalized disturbance and unknown pseudo partial derivative are, respectively, estimated by an ESO and a projection-like algorithm. To further optimize the tracking error, a data driven SMPLC scheme is developed by using the model predictive control approach. Robust convergence is rigorously analyzed by the contraction mapping principle. The effectiveness and superiority of the proposed scheme are demonstrated by an illustrative example.