Dr. Matwin, Here you go. Marc... ------------------------------------------------------------------------------- Marc Seguin Nortel Public Carrier Networks World Trade Technology Tel: (613) 765-3744 email: mseguin@nortel.ca ---forwarded-message----> May 21 13:33 1998 To: Marc Seguin :7Z11 (BNR) SKY BNR From: 'stan@site.uottawa.ca' (INTERNET) Subject: our results (fwd) - re summer project Attached: 1 UNIX File: celtic.label.gz 4560 bytes 2 UNIX File: fundul.label.gz 4720 bytes 3 UNIX File: augusta.label.gz 4560 bytes 4 UNIX File: ORIGINAL.HEADER 960 bytes Stan Matwin School of Information Technology email: stan@csi.uottawa.ca and Engineering (SITE) phone: (613) 562-5800 ext. 6679 University of Ottawa fax: (613) 562-5187 150 Louis Pasteur WWW: http://www.csi.uottawa.ca/~stan Ottawa, Ontario K1N 6N5 Canada ---------- Forwarded message ---------- Date: Tue, 12 May 1998 11:36:24 +0200 (MET DST) From: Andreas Ittner To: Stan Matwin Subject: our results Hallo Stan, here are our results concerning your machine learning task. We've compared C4.5 and our machine learning algorithm. As an attachment you will get the labeled celtic, fundul and augusta files. These files are compressed with gzip. In a pre-processing step we've normalized the data. Class 5 has been mapped to 4, because it was more convenient. We would be very interested in your feedback with respect to our results. May be there are the original class labels of the three files mentioned above and we can also compute the accuracies. Greetings from Chemnitz, Andreas ----------------------------------------------------------------------- Andreas Ittner Phone: +49-371-531-1643 Chemnitz University of Technology Fax: +49-371-531-1465 Department of Computer Science Artificial Intelligence Research Group D-09107 Chemnitz Germany E-Mail: andreas.ittner@informatik.tu-chemnitz.de WWW: http://www.tu-chemnitz.de/~ait/andreas.html ----------------------------------------------------------------------- ******************************************************************* C4.5 results: ------------- learn_x.data: Evaluation on training data (1910 items): Before Pruning After Pruning ---------------- --------------------------- Size Errors Size Errors Estimate 377 69( 3.6%) 295 100( 5.2%) (14.7%) << test_x.dat: Evaluation on test data (40 items): Before Pruning After Pruning ---------------- --------------------------- Size Errors Size Errors Estimate 377 12(30.0%) 295 13(32.5%) (14.7%) << (a) (b) (c) (d) <-classified as ---- ---- ---- ---- 7 2 1 (a): class 1 1 6 2 1 (b): class 2 3 1 6 (c): class 3 2 8 (d): class 4 ******************************************************************* our results: ------------ training errors (learn_x.data): 87 (4.6%) (a) (b) (c) (d) <-classified as ---- ---- ---- ---- 424 3 24 11 (a): class 1 1 470 1 2 (b): class 2 24 5 447 4 (c): class 3 5 7 482 (d): class 4 test errors (test_x.dat): 8 (20.0%) (a) (b) (c) (d) <-classified as ---- ---- ---- ---- 7 1 2 (a): class 1 10 (b): class 2 2 1 6 1 (c): class 3 1 9 (d): class 4 number of examples per class: class 1 2 3 4 ----------------------------------------- celtic.label 79 36 4 fundul.label 49 35 16 19 augusta.label 8 110 2