You must be meaning MOLUSCE: never fully tried it myself but their quick help (accessible from the menu and in the QGIS wiki) is rather explicit on the meaning and contents of the tabs, that should be used sequentially:

• Define your inputs with t-1 and t0 raster files + additional variables (eg. distance to roads, cities, ...)
• Analyse the t-1-t0 transitions with Evaluating correlation and Area Changes
• Build, run and evaluate the LUCC model with Transition Potential Modelling, Cellular Automata Simulation, Validation

There are many pathways for the last step. But MOLUSCE seems to be sharing most of its algorithms with existing frameworks. You might have a look at the Mas & al paper, where they describe and compare in detail the modelling approaches of IDRISI's modules, Dinamica and CLUE-S.  

http://hub.qgis.org/projects/molusce/wiki

Article Inductive pattern-based land use/cover change models: A comp...

Hi Baptiste,

I'm trying to working with molusc qgis too but my problem is the kind of the additional variable. You said cities or distance fron road. There must on point or polygone form with attribute table. I'm working on mangrove change modeling.

Thank you for your help

Hello Baptiste Hautdidier

please enlighten me more on how to obtain spatial variables such as elevation, slope, distance to road and population density which a re inputs to the model.