Modelling urban land use change is of profound concern to environmental scientists who have found cellular automata models very attractive for simulating urban dynamics. The quest for suitable predictive models to improve realistic simulation of urban land use change has resulted in the use of several notable cellular automata calibrations. Cellular automata model has become very attractive and one of the strongest models for urban growth simulation due to its simplicity and possibility of evolution. However, the inability of cellular automata to include driving forces of urban growth in the simulation process has warranted further cellular automata calibrations to minimize this weakness. To address this problem, and contrary to previous cellular automata calibrations, this research presents a novel integration of support vector machine, Markov chain and cellular automata for urban change modelling. Support vector machine is introduced as a machine learning technique to mine the impact of the explanatory variables that drive urban change. Markov chain is employed to mine the urban transition probabilities between the various urban epochs while cellular automata are used to implement the incremental discrete time steps based on neighbourhood interaction from an initial time to a future time. This modelling is implemented using Landsat data acquired in 1984, 2000 and 2015 over Lagos in Nigeria; Africa’s most populous city. Urban transitions (1984-2000 and 2000-2015) are used to simulate future urban state in 2030 and validation metrics include McNemar's test. The introduction of stochasticity into the model helps create the typical randomness inherent in the real world for deriving future urban forms through discrete cellular automata iterations. The high accuracy obtained in this experiment implies a substantial fit between the predicted and reference data, which proves the robustness of this method for modelling urban change.