Overview

This book presents a unified framework, based on specialized evolutionary algorithms, for the global induction of various types of classification and regression trees from data. The resulting univariate or oblique trees are significantly smaller than those produced by standard top-down methods, an aspect that is critical for the interpretation of mined patterns by domain analysts. The approach presented here is extremely flexible and can easily be adapted to specific data mining applications, e.g. cost-sensitive model trees for financial data or multi-test trees for gene expression data. The global induction can be efficiently applied to large-scale data without the need for extraordinary resources. With a simple GPU-based acceleration, datasets composed of millions of instances can be mined in minutes. In the event that the size of the datasets makes the fastest memory computing impossible, the Spark-based implementation on computer clusters, which offers impressive fault tolerance and scalability potential, can be applied.