50 likes | 59 Views
A method combining ensemble of trees and neural networks, ranked 30th at PAKDD 2007. Handles multiple goals efficiently without overfitting. Utilizes k-folds-validation and genetic algorithms to prevent overlearning. Suitable for small category problems and various managerial objectives. Emphasizes testing different models and exploring all possibilities to deliver top solutions.
E N D
Ensemble of ensemble of tree and neural network Louis Duclos-Gosselin
Plan of the presentation • Introduction • General Model • Conclusion
Introduction • This type of model was ranked 30th at PAKDD 2007 • The approach explores all possibility and all kind of model to bring The Best Model • It doesn’t over fit • It can manage multiple managerial goal
Conclusion • The strength of my method is this kind of algorithm doesn’t over fitting because k-folds-validation and genetic algorithms are used during all the process to keep the over learning as low as possible • This process is particular powerful on small category problem • It can handle different managerial goal • All the possibility are explored; all the architecture are visited; all the parameters are tested • We should always test all possibility, all category of model and all new ideas available in the literature to provide The Best Solution to the manager • We must keep us informed about The Best Technique