GATree Genetically Evolved Decision Trees Papagelis Athanasios - Kalles Dimitrios Computer Technology Institute
Introduction We use GA’s to evolve simple and accurate binary decision trees Simple genetic operators over tree structures Experiments with UCI datasets very good size competitive accuracy results
Why it should work ? GA’s are not They are … Hill climbers Blind on complex search spaces Exhaustive searchers Extremely expensive They are … Beam searchers They balance between time needed and space searched
The question… Are there datasets where hill-climbing techniques are really inadequate ? e.g unnecessary big – misguiding output Yes there are… Conditionally dependent attributes e.g XOR Irrelevant attributes Many solutions that use GAs as a preprocessor so as to select adequate attributes Direct genetic search can be proven more efficient for those datasets
The proposed solution Select the desired decision tree characteristics (e.g small size) Create an appropriate fitness function Adopt a decision tree representation with appropriate genetic operators Evolve for as long as you wish!
Genetic operators
Payoff function Balance between accuracy and size set x depending on the desired output characteristics. Small Trees ? x near one Emphasis on accuracy ? x grows big
Results
Future work Minimize evolution time Improved node statistics Choose the output class using a majority vote over the produced tree forest Dynamic tuning of initial parameters Experiments with synthetic datasets Specific characteristics