Steps of capturing patterns from data called fitting or training. The data used to fit/train the model called the training data.
Then after the model has been fit, we apply it to new data to predict
In decision tree, the last bottom point where the prediction placed called a leaf
Mean Absolute Error(MAE) counts from 0 to infinity, it means this evaluation metrics cannot be used as a standalone evaluation metrics without context. It needs domain knowledge or business context for the score have a meaning. The stakeholders must know the expected value scale based on the business they have.
MAE more suitable used for comparing different machine learning models on the same dataset, to find the better model by looking at the lowest score between models.
MAE can also be used for comparing values of training data and testing/validation data during evaluation, to estimate the underfitting and overfitting, visually using linear graph.