source: orange/orange/doc/datasets/ionosphere.htm @ 1760:9d4bb141fb0e

Revision 1760:9d4bb141fb0e, 3.2 KB checked in by blaz <blaz.zupan@…>, 9 years ago (diff)

data info file

Line 
1<html>
2<head>
3<title>Johns Hopkins University Ionosphere Data Base</title>
4</head>
5<body>
6<h1>Info on Johns Hopkins University Ionosphere Data Base</h1>
7<pre>
81. Title: Johns Hopkins University Ionosphere database
9
102. Source Information:
11   -- Donor: Vince Sigillito (vgs@aplcen.apl.jhu.edu)
12   -- Date: 1989
13   -- Source: Space Physics Group
14              Applied Physics Laboratory
15              Johns Hopkins University
16              Johns Hopkins Road
17              Laurel, MD 20723
18
193. Past Usage:
20   -- Sigillito, V. G., Wing, S. P., Hutton, L. V., \& Baker, K. B. (1989).
21      Classification of radar returns from the ionosphere using neural
22      networks. Johns Hopkins APL Technical Digest, 10, 262-266.
23
24      They investigated using backprop and the perceptron training algorithm
25      on this database.  Using the first 200 instances for training, which
26      were carefully split almost 50% positive and 50% negative, they found
27      that a "linear" perceptron attained 90.7%, a "non-linear" perceptron
28      attained 92%, and backprop an average of over 96% accuracy on the
29      remaining 150 test instances, consisting of 123 "good" and only 24 "bad"
30      instances.  (There was a counting error or some mistake somewhere; there
31      are a total of 351 rather than 350 instances in this domain.) Accuracy
32      on "good" instances was much higher than for "bad" instances.  Backprop
33      was tested with several different numbers of hidden units (in [0,15])
34      and incremental results were also reported (corresponding to how well
35      the different variants of backprop did after a periodic number of
36      epochs).
37
38      David Aha (aha@ics.uci.edu) briefly investigated this database.
39      He found that nearest neighbor attains an accuracy of 92.1%, that
40      Ross Quinlan's C4 algorithm attains 94.0% (no windowing), and that
41      IB3 (Aha \& Kibler, IJCAI-1989) attained 96.7% (parameter settings:
42      70% and 80% for acceptance and dropping respectively).
43
444. Relevant Information:
45   This radar data was collected by a system in Goose Bay, Labrador.  This
46   system consists of a phased array of 16 high-frequency antennas with a
47   total transmitted power on the order of 6.4 kilowatts.  See the paper
48   for more details.  The targets were free electrons in the ionosphere.
49   "Good" radar returns are those showing evidence of some type of structure
50   in the ionosphere.  "Bad" returns are those that do not; their signals pass
51   through the ionosphere. 
52
53   Received signals were processed using an autocorrelation function whose
54   arguments are the time of a pulse and the pulse number.  There were 17
55   pulse numbers for the Goose Bay system.  Instances in this databse are
56   described by 2 attributes per pulse number, corresponding to the complex
57   values returned by the function resulting from the complex electromagnetic
58   signal.
59
605. Number of Instances: 351
61
626. Number of Attributes: 34 plus the class attribute
63   -- All 34 predictor attributes are continuous
64
657. Attribute Information:     
66   -- All 34 are continuous, as described above
67   -- The 35th attribute is either "good" or "bad" according to the definition
68      summarized above.  This is a binary classification task.
69
708. Missing Values: None
71</pre>
72</body>
73</html>
Note: See TracBrowser for help on using the repository browser.