Splitting criteria are decision rules used in decision trees to determine how to split data at each node, aiming to increase the homogeneity of the resulting subsets. Effective splitting criteria are crucial for building accurate and efficient decision tree models by minimizing impurity and maximizing information gain.