|
Abstract:
|
In any model-building process, relationships between predictor variables cause complications. In "feature selection" approaches such as recursive partitioning (RP), correlated predictors may be mutually substitutable, make additive contributions, or show synergy. These features are rarely discovered in recursive partitioning models because most RP methods generate a single tree as "the" answer. The solution is to produce not a single tree, but a "forest" of models that fit the data. Harvesting knowledge from the forest is much harder than generating it though. Not only may visually different trees be identical in the sense of generating identical rules, but given correlated predictors, different rules may lead to identical sets of terminal nodes. Different methods of building a "forest" of tree-based models will be discussed and applied to gene expression data.
|
