Project 10.2: Neural Networks Revisted : Report

Intro ] [ codedescription ] [ final-summary ] [ summary ]

Experiments ] [ Up: Neural Networks Revisited ]

Technique Decision Trees Neural Networks Naive Bayes Instance Based Genetic Algorithms Rule Learning Final: Neural Networks
Best accuracy 86.0881 77.1329 84.2700 81.3893 76.8932 81.3341 84.3806
Code (mine/other/adapted) Weka Adapted from Book Adapted Weka Adapted Weka (nearest neighbor) Mine (and it shows) FOIL (Quinlan) Adapted from previous
Programming Language Java Java Java Java Java C Java
Num. of training instances 32561 500 32561 32561 515 32561 32561
Num. of test instances 16281 16281 16281 16281 16281 16281 16281
Missing values included?(y/n) Yes Yes Yes Yes Yes Yes Yes
Number of attributes used All All All All All All All
Best thing about this project -Good intro -Tree display -Cool sounding -Expect good accuracy -Writing it -Simple concept -Simple code -Good initial results -Extended code -Case based reasoning -Comparable - Weka's IBk -Tried novel approach -Discussion despite results -Impressive system -Difficult by hand -Excellent readability -Verified functionality -Accuracy better -Complex behavior fun
Worst thing about this project -Lack expectations -Didn't know if accuracy was good -Poor accuracy -Verifying functionality -Tuning less apparent -Explaining Std. Dev., mean for num. att. -Time lacking : more tests -Poor results -Uncompleted (short-sighted) system vision -Jargon ignorance = confusion -No Java port time ;) -Run time -Complex behavior (no play time)
Other comments Compare tree and FOIL rules Needed save/load Still need save/load
 

by: Keith A. Pray
Last Modified: July 4, 2004 9:05 AM
© 2004 - 1975 Keith A. Pray.
All rights reserved.