Ensemble-learning-based Cellular Automata for Urban Geosimulation

This paper presents a new method to simulate urban dynamics based on ensemble learning, cellular automata and GIS. Ensemble learning is an important direction of the machine learning which can improve the performance than the single classifier by constructing a set of classifier and then classify the new data set by taking a weighted vote of their predictions. Many algorithms have been used to simulate the change of land use so far. However, there exists some limitations about the algorithms in the applications . For instance, the dicision trees algorithm is always confronting the problem of over-fitting, the neural network algorithm would be trapped in local optimum easily. This paper uses the dicision tree algorithm as the classifier of the ensemble learning to predict urban dynamics of Dongguan city. The result shows that the simulation of the ensemble learning is better than the decision tree.