github.com/vlifesystems/rulehunter@v0.0.0-20180501090014-673078aa4a83/examples/csv/breast_cancer_wisconsin.txt (about) 1 1. Title: Wisconsin Diagnostic Breast Cancer (WDBC) 2 3 2. Source Information 4 5 a) Creators: 6 7 Dr. William H. Wolberg, General Surgery Dept., University of 8 Wisconsin, Clinical Sciences Center, Madison, WI 53792 9 wolberg@eagle.surgery.wisc.edu 10 11 W. Nick Street, Computer Sciences Dept., University of 12 Wisconsin, 1210 West Dayton St., Madison, WI 53706 13 street@cs.wisc.edu 608-262-6619 14 15 Olvi L. Mangasarian, Computer Sciences Dept., University of 16 Wisconsin, 1210 West Dayton St., Madison, WI 53706 17 olvi@cs.wisc.edu 18 19 b) Donor: Nick Street 20 21 c) Date: November 1995 22 23 Data Source: 24 UCI Machine Learning Repository [http://archive.ics.uci.edu/ml]. 25 Irvine, CA: University of California, School of Information and 26 Computer Science. 27 https://archive.ics.uci.edu/ml/datasets/Breast+Cancer+Wisconsin+(Diagnostic) 28 29 3. Past Usage: 30 31 first usage: 32 33 W.N. Street, W.H. Wolberg and O.L. Mangasarian 34 Nuclear feature extraction for breast tumor diagnosis. 35 IS&T/SPIE 1993 International Symposium on Electronic Imaging: Science 36 and Technology, volume 1905, pages 861-870, San Jose, CA, 1993. 37 38 OR literature: 39 40 O.L. Mangasarian, W.N. Street and W.H. Wolberg. 41 Breast cancer diagnosis and prognosis via linear programming. 42 Operations Research, 43(4), pages 570-577, July-August 1995. 43 44 Medical literature: 45 46 W.H. Wolberg, W.N. Street, and O.L. Mangasarian. 47 Machine learning techniques to diagnose breast cancer from 48 fine-needle aspirates. 49 Cancer Letters 77 (1994) 163-171. 50 51 W.H. Wolberg, W.N. Street, and O.L. Mangasarian. 52 Image analysis and machine learning applied to breast cancer 53 diagnosis and prognosis. 54 Analytical and Quantitative Cytology and Histology, Vol. 17 55 No. 2, pages 77-87, April 1995. 56 57 W.H. Wolberg, W.N. Street, D.M. Heisey, and O.L. Mangasarian. 58 Computerized breast cancer diagnosis and prognosis from fine 59 needle aspirates. 60 Archives of Surgery 1995;130:511-516. 61 62 W.H. Wolberg, W.N. Street, D.M. Heisey, and O.L. Mangasarian. 63 Computer-derived nuclear features distinguish malignant from 64 benign breast cytology. 65 Human Pathology, 26:792--796, 1995. 66 67 See also: 68 http://www.cs.wisc.edu/~olvi/uwmp/mpml.html 69 http://www.cs.wisc.edu/~olvi/uwmp/cancer.html 70 71 Results: 72 73 - predicting field 2, diagnosis: B = benign, M = malignant 74 - sets are linearly separable using all 30 input features 75 - best predictive accuracy obtained using one separating plane 76 in the 3-D space of Worst Area, Worst Smoothness and 77 Mean Texture. Estimated accuracy 97.5% using repeated 78 10-fold crossvalidations. Classifier has correctly 79 diagnosed 176 consecutive new patients as of November 80 1995. 81 82 4. Relevant information 83 84 Features are computed from a digitized image of a fine needle 85 aspirate (FNA) of a breast mass. They describe 86 characteristics of the cell nuclei present in the image. 87 A few of the images can be found at 88 http://www.cs.wisc.edu/~street/images/ 89 90 Separating plane described above was obtained using 91 Multisurface Method-Tree (MSM-T) [K. P. Bennett, "Decision Tree 92 Construction Via Linear Programming." Proceedings of the 4th 93 Midwest Artificial Intelligence and Cognitive Science Society, 94 pp. 97-101, 1992], a classification method which uses linear 95 programming to construct a decision tree. Relevant features 96 were selected using an exhaustive search in the space of 1-4 97 features and 1-3 separating planes. 98 99 The actual linear program used to obtain the separating plane 100 in the 3-dimensional space is that described in: 101 [K. P. Bennett and O. L. Mangasarian: "Robust Linear 102 Programming Discrimination of Two Linearly Inseparable Sets", 103 Optimization Methods and Software 1, 1992, 23-34]. 104 105 106 This database is also available through the UW CS ftp server: 107 108 ftp ftp.cs.wisc.edu 109 cd math-prog/cpo-dataset/machine-learn/WDBC/ 110 111 5. Number of instances: 569 112 113 6. Number of attributes: 32 (ID, diagnosis, 30 real-valued input features) 114 115 7. Attribute information 116 117 1) ID number 118 2) Diagnosis (M = malignant, B = benign) 119 3-32) 120 121 Ten real-valued features are computed for each cell nucleus: 122 123 a) radius (mean of distances from center to points on the perimeter) 124 b) texture (standard deviation of gray-scale values) 125 c) perimeter 126 d) area 127 e) smoothness (local variation in radius lengths) 128 f) compactness (perimeter^2 / area - 1.0) 129 g) concavity (severity of concave portions of the contour) 130 h) concave points (number of concave portions of the contour) 131 i) symmetry 132 j) fractal dimension ("coastline approximation" - 1) 133 134 Several of the papers listed above contain detailed descriptions of 135 how these features are computed. 136 137 The mean, standard error, and "worst" or largest (mean of the three 138 largest values) of these features were computed for each image, 139 resulting in 30 features. For instance, field 3 is Mean Radius, field 140 13 is Radius SE, field 23 is Worst Radius. 141 142 All feature values are recoded with four significant digits. 143 144 8. Missing attribute values: none 145 146 9. Class distribution: 357 benign, 212 malignant