Int J Pharm Pharm Sci, Vol 6, Issue 8, 117-122Original Article

ENDORSEMENT OF SMALL PATIENTS POPULATION STUDY THROUGH DATA MINING CLASSIFICATION: SIGNIFICANCE TO MANIFEST DRUG INTERACTION STUDY OF CARDIOVASCULAR DOSAGE FORMULATION

RAKESH DAS, SUBHASIS DAN, TAPAN KUMAR PAL*

Bioequivalence Study Center, Dept. of Pharmaceutical Technology, Jadavpur University, Jadavpur, Kolkata, W.B. India
Email: tkpal12@gmail.com

Received: 19 Jun 2014 Revised and Accepted: 22 Jul 2014

ABSTRACT

Objective: A simple, sensitive, precise computational classifiers justifies the positive indication of drug interaction through statistical validation and confirms for further root level investigation.

Methods: The blood pressure (BP) & Lipid profile valued data sheet was prepared from 100 patients those were chronically treating with cardiovascular formulation consisting Atorvastatin 10mg + Olmesartan 20mg. The data sheet contains 100 patients with 10 variables and final decision attributes of working & non-working. Then, with the operation of seven different related classifier the details of % of accuracy by class, correct & incorrect classified instance and stratified cross- validation were estimated. Those statistical results of classifiers were compared, correlate and interpreted to bring a fixed conclusion based on it.

Results: The % of accuracy for all classifiers results commonly 95.9596 %, 93.9394 % and 96.9697 % and inter-depending class attributes denoting by a = NW & b =W Matrix values are 84│11, 84│9, 87│9 respectively. Thus, the accuracy is excellent covering within the limits of (±15%) as a correct classified instant.

Conclusion: Statistical computation on less populated patients through classifiers, evidentially confirms the drug-interaction profile of collected data through data mining process. So that, it can proceeds further upto root level through instrumental bioanalysis.

Keywords: Seven Classifiers, computatIonal statistical analysis, Physio-chemical patients data, Data mining process.

INTRODUCTION

Rapid inter-collaboration between Clinical pharmacy researchers working in Cardiovascular therapy and computer scientists are looking at the application of data mining techniques to the area of individual patients diagnosis, based on the clinical records. An investigation of seven different classification models on cardiovascular data for estimation of patient risk in cardiovascular domains is presented [1-7].

A major Challenge on healthcare sectors is the provision of quality services at affordable cost. Quality service implies diagnosing patients exactly and providing treatments that are cost-effective. Bad clinical decision can lead to negative consequences which are therefore wasteful. They can get onto these results by implicating appropriate computer-based information and /or decision support systems [8]. One in eight women over their lifetime has a risk of developing breast cancer. An analysis of the most recent data has shown that the survival rate is 88% after 5 years of diagnosis and 80% after 10 years of diagnosis [9]. The nature of a population can be seemed to establish the reasons associated with a specific endpoints. Prospective studies, such as statistical learning and data mining, can approach the association of the variables to the outcome, but were not always establish the cause-and-effect relationship of the association. Data holding statistical research is becoming a common breeze to many scientific areas like medicine and biotechnology. This trend commonly observed as in the studies of Houston et al. and Cios et al. [10-11]. A literature survey convey several studies on the survivability prediction problem using statistical approaches and artificial neural networks. However, we observe few studies related to medical diagnosis and survivability using data mining approaches like decision trees [12-14]. Other than the breadth of stored information, which gradually includes long-term outcome and associated biological and genetic data, mining for potentially novel and useful biomedical associations in Clinical Data Repositories (CDRs) is a relatively recent approach [15-18]. Research has dual effect: to develop clinical participated databases of cancer patients, and to conduct data mining and learning studies on collected patient records [19]. Statistics and data mining differ in the use of machine learning methods, the volume of data, and the role of computational complexity. Requirement for analysis is preceding our abilities to handle the complexity. Preprocessing is much vital with large datasets, especially as we reaches the pentabyte level. However, data mining is concentrate on the data mining process itself with little traces on the knowledge actually extracted[20-22]. Cardiovascular decision-making support experiences increasing research interest of scientists. Simultaneous collaborations between clinical pharmacy researchers and computer scientists are focusing at the implication of data mining techniques to the area of individual patient diagnosis, based on clinical datas, Bayesian network [23].

MATERIALS AND METHODS

Study Design

On evolving negative therapeutic information for Cardiovascular combined drug formulation of Atorvastatin (10mg) & Olmesartan (20mg) from patient’s Clinical history, doctors comments, mortility & morbidity frequencies, clinical study, which was operated on Midinapur Medical college and Hospital under clinical supervisor Dr. Balaram Ghosh. The prospective observational study were conducted among the 100 patients those were under chronic therapy of that formulation. The basic data were collected on the basis of Blood pressure (BP), Lipid profiles (HDL, LDL, VLDL, Triglycerides, Total Cholesterol, Total- Cholest/ HDL, LDL-C/HDL-C levels) and there hypothetical correlative resulting data, Working (W) & Non-working (NW) from patients physiological and biochemistry data chart reports.

Data assemblance

Systolic BP & Diastolic BP of 100 patients were assembled in successively in 1st and 2nd column. The lipid profile for those corresponding 100 patients were spilt to HDL, LDL, VLDL, Triglyceride in chronicle manners in columns, followed by a depending column of Total Cholesterol. The 3rd last and 2nd last column assists the functional ratio of T-Cholesterol/HDL & LDL-C/HDL-C respectively. Also, in final column attributes are domain with working (W) and Non- working (NW). This patients data sheet were included in Table- 1.

Table 1: The medical data sheet exhibiting BP & Lipid profile collected after chronic administration of CVD formulation of 100 patients.

Systolic BPDiastolic BPHDLLDLVLDLTriglycerideT.CholT.Chol/HDLLDL-C/HDL-Cclass
172 98 39 143 43 212 198 5.09 3.66 N.W
176 123 36 162 46 231 215 5.97 4.5 N.W
169 136 40 169 42 216 223 5.57 4.22 N.W
173 125 39 155 46 198 234 6 3.97 N.W
168 138 40 159 43 219 197 4.92 3.97 N.W
170 141 36 162 48 223 231 6.41 4.5 N.W
173 139 32 159 46 241 224 7 4.96 N.W
189 141 37 160 42 214 231 6.24 4.32 N.W
172 138 41 163 41 209 214 5.21 3.97 N.W
168 132 35 141 46 213 235 6.71 4.02 N.W
186 134 59 143 46 200 237 4.01 2.42 N.W
180 121 32 157 43 212 224 7 4.9 N.W
167 140 36 146 45 198 232 6.44 4 N.W
182 156 63 144 41 159 214 3.39 2.28 N.W
168 141 36 172 48 172 242 6.72 4.77 N.W
165 100 44 164 42 198 232 5.27 3.72 N.W
177 126 23 139 46 199 219 9.52 6.04 N.W
152 135 69 163 47 222 221 3.2 2.36 N.W
163 132 71 167 42 256 251 3.53 2.35 N.W
178 133 37 143 48 159 236 6.37 3.86 N.W
173 145 43 129 50 168 228 5.3 3 N.W
188 134 41 161 53 183 194 4.73 3.92 N.W
169 123 79 158 42 199 233 2.94 2 N.W
154 119 38 159 51 198 246 6.47 4.18 N.W
183 141 36 165 46 189 213 5.91 4.58 N.W
174 141 69 157 47 231 239 3.46 2.27 N.W
167 134 72 147 43 212 227 3.15 2.04 N.W
124 102 29 112 27 156 195 6.72 3.86 W
183 143 37 134 46 126 249 6.72 3.62 N.W
168 135 35 167 48 147 226 6.45 4.77 N.W
170 142 32 163 44 198 245 7.65 5.09 N.W
189 150 33 158 72 242 243 7.36 4.78 N.W
202 153 33 174 58 271 225 6.81 5.27 N.W
154 96 41 159 54 233 245 5.97 3.87 N.W
164 136 47 169 61 213 236 5.02 3.59 N.W
187 143 40 163 45 243 235 5.87 4.07 N.W
178 126 39 145 43 232 237 6.07 3.71 N.W
167 129 78 160 47 251 231 2.96 2.05 N.W
153 142 29 147 53 202 221 7.62 5.06 N.W
131 123 33 102 30 123 198 6 3.09 N.W
127 89 42 100 27 134 200 4.76 2.38 N.W
112 91 62 95 25 128 187 3.01 1.53 W
190 141 69 170 61 198 236 3.42 2.46 N.W
162 132 60 145 46 231 244 4.06 2.41 N.W
176 132 28 163 49 231 238 8.5 5.82 N.W
168 121 36 156 44 212 237 6.58 4.33 N.W
171 124 38 154 39 189 247 6.5 4.05 N.W
166 112 35 155 38 199 239 6.82 4.42 N.W
182 142 45 170 51 213 235 5.22 3.77 N.W
175 145 42 182 46 215 231 5.5 4.33 N.W
159 120 34 156 57 232 234 6.88 4.58 N.W
169 112 40 149 59 230 241 6.02 3.72 N.W
182 131 72 157 65 219 234 3.25 2.18 N.W
117 82 37 98 32 141 197 5.32 2.64 W
174 147 39 145 63 199 216 5.53 3.71 N.W
165 125 43 134 57 250 234 5.44 3.11 N.W
184 136 47 147 48 262 267 5.68 3.12 N.W
155 114 37 134 52 202 241 6.51 3.62 N.W
162 118 49 139 38 210 242 4.93 2.83 N.W
139 100 68 154 58 223 211 3.1 2.26 N.W
173 152 38 137 49 232 239 6.28 3.6 N.W
120 78 45 98 28 135 187 4.15 2.17 W
181 135 40 126 39 209 242 6.05 3.15 N.W
118 85 32 100 25 129 176 5.5 3.12 W
165 140 38 145 58 231 241 6.34 3.81 N.W
168 131 23 154 55 214 227 9.86 6.69 N.W
174 127 25 171 62 235 238 9.52 6.84 N.W
186 132 39 137 48 223 229 5.87 3.51 N.W
159 89 55 148 67 231 235 4.27 2.69 N.W
165 152 21 187 73 235 223 10.61 8.9 N.W
158 102 70 169 39 227 242 3.45 2.41 N.W
168 115 50 135 54 222 225 4.5 2.7 N.W
174 91 20 165 46 198 237 11.85 8.25 N.W
176 119 42 155 48 199 226 5.38 3.69 N.W
179 138 24 143 54 216 256 10.66 5.95 N.W
142 111 21 138 81 189 249 11.85 6.57 N.W
197 152 47 154 46 217 265 5.63 3.27 N.W
160 129 49 149 39 208 243 4.95 3.04 N.W
123 82 46 101 32 142 179 3.89 2.19 W
115 74 53 98 21 126 181 3.41 1.84 W
127 68 22 96 26 137 189 8.59 4.36 W
176 124 23 156 56 214 265 11.52 6.78 N.W
178 131 22 147 47 207 246 11.18 6.68 N.W
183 115 47 154 54 200 243 5.17 3.27 N.W
155 103 76 147 38 213 233 3.06 1.93 N.W
171 131 65 157 68 215 255 3.92 2.41 N.W
116 74 48 153 54 142 188 3.91 3.18 N.W
121 78 37 98 30 138 185 5 2.64 W
172 116 20 158 45 215 232 11.6 7.9 N.W
189 149 58 154 76 223 265 4.56 2.65 N.W
178 126 72 148 69 200 234 3.37 2.05 N.W
125 80 54 100 23 157 190 3.51 1.85 W
181 152 23 158 46 214 216 9.39 6.86 N.W
125 78 19 92 19 148 194 10.21 4.84 N.W
162 112 24 154 54 231 235 9.79 6.41 N.W
152 100 75 146 48 225 247 3.29 1.94 N.W
187 102 43 148 54 218 243 5.65 3.44 N.W
123 78 43 99 24 132 198 4.6 2.3 W
120 79 56 89 26 136 202 3.6 1.58 W

Foot Notes: Blood Pressure, BP; High Density Lipoprotein, HDL; Low Density Lipoprotein, LDL; Very Low Density Lipoprotein, VLDL; Cholesterol, C; Total Cholesterol, T.Chol; Not working, NW; Working, W; Cardiovascular drug, CVD.

Computational Analysis

Computational analysis were carried out through various important workable and relating classifiers i.e, a) NaiveBayes; b) SMO; c) Lazy.KStar Beta-version; d) Meta. adaBoostM1; e) Meta. Bagging; f) rules. PART; and g) Tress. J48, to understand the percentage (%) of accuracy analysis in those small population of 100 patients. Based on results to these analytical justification through those classifiers, the further permission and requirement of drug interaction study proceedings could be determined.

Accuracy evaluation

The weka software is most supportive to compute data sheet variables. Each variables of all the patients were statistically developed on respect to Mean, Std. dev., weight sum, & precision. So that according to each & every seven classifier, accuracy could be traced out. These accuracy and inexactness ratio of all well reputed classifiers reflects its signified conclusion after correlating. The standard statistical reports would aid on making decision to prepare valid reason to start investigation even on lower populated patients data.

Recruitment of based Analysis

The analysis of bio-analytes (peptides, enzymes & other biochemical product) and analytes inter-related to investigation content is required to access drug-interactions of patients out of 100 enrolled in data sheet.

RESULTS

The biochemical and physiological variation marked after the cardiovascular drug therapy among all 100 patients, were statistically analyse for the accuracy using computational process of data mining, Figure-1.

The computational statistical complete analysis were carried out with classifiers- Naive Bayes Classifier, SMO, KStar Beta Verion (0.1b), AdaBoost, bagging, rules. PART; and Trees. J48.

Fig. 1: Classification of accuracy with best data mining tool classifiers.


Table 2: Statistical results of all correctly classified (N.W) and incorrectly classified (N) attribute for BP and Lipid profile analysis as per model classifiers.

ClassAttributeAttribute
N.W W
(0.87) (0.13)
Systolic BP
mean 169.6552 120.45
std. dev. 15.0777 4.1866
weight sum 87 12
precision 1.8 1.8
Diastolic BP
mean 127.0179 81.6028
std. dev. 17.6302 8.0336
weight sum 87 12
precision 1.8723 1.8723
HDL
mean 43.0135 42.9348
std. dev. 15.5519 11.5989
weight sum 87 12
precision 1.3043 1.3043
LDL
mean 152.2943 98.5444
std. dev. 15.4126 4.715
weight sum 87 12
precision 2.1778 2.1778
VLDL
mean 49.4679 26.7154
std. dev. 10.1452 3.2276
weight sum 87 12
precision 1.5122 1.5122
Triglyceride
mean 209.4661 137.761
std. dev. 28.169 9.5607
weight sum 87 12
precision 2.7925 2.7925
T.Cholesterol
mean 232.4631 188.9643
std. dev. 15.9614 7.8472
weight sum 87 12
precision 1.8571 1.8571
T.Chol/HDL
mean 6.1322 4.772
std. dev. 2.2604 1.5425
weight sum 87 12
precision 0.1001 0.1001
LDL-C/HDL-C
mean 3.996 2.5091
std. dev. 1.5216 0.8554
weight sum 87 12
precision 0.0899 0.0899

Results of the cross-validation were justified statistically under Naive Bayes, SMO, AdaBoost, Bagging Classifiers. The classified instant which carried 95.95% of correctly and 4.04% of incorrect accuracies. Other than that Naive Bayes has kappa statistic, Mean absolute error, root mean squared error, relative absolute error and Root relative squared error are 0.8231, 0.0398, 0.1983, 18.13%, 60.62% respectively. And the population of instances on patients investigation was performed out-off 100. Also, the not working (NW) & working (W) classes were expressed in 84│3│ and 1│11 in matrix calculation respectively for Naïve Bayes classifier. According to SMO classifier, number of kernel evaluations are 322 (69.594% cached). Statistical calculations of SMO; AdaBoost; Bagging classifiers upon kappa statistic, Mean absolute error, root mean squared error, relative absolute error and Root relative squared error are exhibits 0.8242, 0.0404, 0.201, 18.39%, 61.46%; 0.8231, 0.043, 0.2015, 19.57%, 61.61%; 0.8231, 0.0813, 0.1925, 36.9984%, 58.8542% respectively. Not working (NW) & working (W) classes were expressed in 84│3│ and 0│12 in matrix calculation respectively for SMO classifier. And not working (NW) & working (W) classes were expressed in 84 │3│ and 1│11 in matrix calculation respectively for AdaBoost & Bagging classifiers.

The cross-validation were justified statistically under Kenstar beta version classifier, were The classified instant which carried 93.93% of correctly and 6.06% of incorrect accuracies. Statistical evaluation of Kenstar beta version classifier lodges kappa statistic, Mean absolute error, root mean squared error, relative absolute error and Root relative squared error are represents- 0.7155, 0.0584, 0.2339, 26.599%, 71.51%. And not working (NW) & working (W) classes were expressed in 84│3│ and 3│9 in matrix calculation respectively for KenSTAR Beta version classifier.

The cross-validation were justified statistically under Rules PART & Trees J48 classifier, were The classified instant which carried 96.9697% of correctly and 3.0303% of incorrect accuracies.

Statistical evaluation of Rules PART & Trees J48 classifiers represents kappa statistic, Mean absolute error, root mean squared error, relative absolute error and Root relative squared error are represents- 0.8406, 0.0303, 0.146, 13.7942%, 44.65% & 0.8406, 0.0303, 0.146, 13.7942%, 44.65% respectively. And not working (NW) & working (W) classes were expressed in 87│0│ and 3│9 in matrix calculation respectively for Rules PART & Trees J48 classifier.

Table 3: Detailed accuracy by class of all 100 patients according to their respective attributes and weighted averages of each status respect to Classifiers.

ClassifiersTP RateFP RatePrecisionRecallF-MeasureROC AreaClass
Naive Bayes 0.966 0.083 0.988 0.966 0.977 0.993 N.W
0.917 0.034 0.786 0.917 0.846 0.993 W
0.96 0.077 0.964 0.96 0.961 0.993 wt. avg.
SMO 0.954 0 1 0.954 0.976 0.977 N.W
1 0.046 0.75 1 0.857 0.977 W
0.96 0.006 0.97 0.96 0.962 0.977 wt.avg.
KenSTAR Beta version 0.966 0.25 0.966 0.966 0.966 0.971 N.W
0.75 0.034 0.75 0.75 0.75 0.971 W
0.939 0.224 0.939 0.939 0.939 0.971 Wt.avg.
Meta AdaBoost 0.966 0.083 0.988 0.966 0.977 0.969 N.W
0.917 0.034 0.786 0.917 0.846 0.969 W
0.96 0.077 0.964 0.96 0.961 0.969 Wt.avg.
Meta Bagging 0.966 0.083 0.988 0.966 0.977 0.97 N.W
0.917 0.034 0.786 0.917 0.846 0.97 W
0.96 0.077 0.964 0.96 0.961 0.97 Wt.avg.
Rules PART 1 0.25 0.967 1 0.983 0.955 N.W
0.75 0 1 0.75 0.857 0.955 W
0.97 0.22 0.971 0.97 0.968 0.955 Wt.avg.
Trees J48 1 0.25 0.967 1 0.983 0.955 N.W
0.75 0 1 0.75 0.857 0.955 W
0.97 0.22 0.971 0.97 0.968 0.955 Wt.avg.

DISCUSSION

On stratified cross-validation after 0.02 sec build model of classifiers - Naive Bayes Classifier, SMO, KStar Beta Verion (0.1b), AdaBoost, bagging, rules. PART; and Trees. J48 represents % of accuracy of 95.9596 %, 95.9596 %, 93.9394 %, 95.9596 %, 95.9596 %, 96.9697 %, 96.9697 % respectively. Also, the incorrect classified instants/ inexactness evolved 4.0404 %, 4.0404 %, 6.0606 %, 4.0404 %, 4.0404 %, 3.0303 %, 3.0303 % respectively. Thus, the accuracy is excellent covering within the limits of (±15%) as a correct classified instant. And depending on the correct classified instant 95, 93, 96 the class attributes denoting by a = NW & b =W Matrix values are 84│11, 84│9, 87│9 respectively. On briefly undergoing through the Kappa Statistics, Mean absolute errors, Root mean squared error, Relative absolute error, root relative square error; and also understanding the correct & incorrect classified instant, it was discovered that classifiers statistical results and values are nearly same.

ABBREVIATIONS

ROC area- Receiver Operating Characteristics; TP-rate- True Positive rate; FP-rate- False Positive rate; F-measure- F-score in F-test measurement; W-working; NW-Not working.

CONCLUSION

The accuracy details, stratified cross-validations and (a,b) matrix of class attributes represents that the statistical results of all the seven classifiers are moreover equal. And its conclude, that data mining application could evaluate and confirms the perfectness of drug interaction even from small populated patients with detail medical history data. And thus, it signifies the positive indication to start investigation further study to get the concrete explanation.

CONFLICT OF INTEREST

We authors are declaring no conflict of interest.

ACKNOWLEDMENT

I myself Mr. Rakesh Das, as a main author of research project funded by All India Council of Technical Education (A.I.C.T.E.) and author Shubhasis Dan is acknowledging University Grants Commission (UGC) for financial support through UGC-BSR fellowship scheme. We authors were extremely thankful to Dr. Balaram Ghosh, MD, Clinical Pharmacologist, Midinapore Medical college and Hospital, West Bengal, to supervising & cooperating to collect medical data reports. And also, I am thankful to patients and patient’s party to allow me for performing uninterrupted research study. I am extremely thankful to faculties of computer Sc. & Engg. for technical support.

REFERENCES

  1. Baesens B, Egmont-Petersen M, Castelo R, Vanthienen J. Learning Bayesian network classifier for credit scoring using Markov chain Monte Carlo serach Proceedings. Int Congress on Pattern Recognition 2002;3:49-52.
  2. Brent M. Instance-Based learning:Nearest neighbor with generalization. Master’s thesis at the university of Waikato, New Zealand;1995. p. 1-76.
  3. Davis DN, Nguyen TT. Generating and verifying risk prediction models using data mining:A case studyfrom Cardiovascular medicine. Chapter of data mining and medical knowledge management:Cases and applications, ISBN10:1605662186. J IGI Global Inc 2009.
  4. Fayyad U, Piatetsky-Shapiro G, Smyth P. From data mining to knowledge discovery in databases. J Al Magazine 1996;17(3):37-45.
  5. Garofalakis M, Hyun D., Rastogi R, Shim K. Building decision trees with constraints. J Data Mining and Knowledge Discovery 2003;7(2):187-214.
  6. Mitchell T M. Machine learning. Mc Graw-Hill Companies. In USA 1997;414.
  7. Nilson NJ. Introduction to machine learning. Unpublished draft;In Standford University, USA, 1996.
  8. Palaniappan S, Awang R. Intelligent heart Disease prediction system using data mining Techniques. Int J of Computer Sc and Network Security 2008;8(8):343-50.
  9. American Cancer Society. Breast Cancer Facts & Figures 2005-2006. Atlanta:American Cancer Society. J Inc
  10. Houston, Andrea L. and Chen, et al. Medical Data Mining on the Internet:Research on a Cancer Information System. J Artificial Intelligence Rev 1999;13:437-66.
  11. Cios KJ, Moore GW. Uniqueness of medical data mining. J Artificial Intelligence in Medicine 2002;26:1-24.
  12. Zhou ZH, Jiang Y. Medical diagnosis with C4.5Rule preceded by artificial neural network ensemble. J IEEE Trans Inf Technol Biomed 2003;7(1):37-42.
  13. Lundin M, Lundin J, Burke HB, Toikkanen S, Pylkkanen L, Joensuu H. Artificial neural networks applied to survival prediction in breast cancer. J Oncology 1999;57: 281-6.
  14. Delen D, Walker G, Kadam A. Predicting breast cancer survivability: a comparison of three data mining methods. J Artificial Intelligence in Medicine 2005;34(2):113-27.
  15. Holmes JH, Durbin DR, Winston FK. Discovery of predictive models in an injury surveillance database: An application of data mining in clinical research. J Proc AMIA Symp 2000;359-63.
  16. Downs SM, Wallace MY. Mining Association rules from a pediatric primary care decision support system. J Proc AMIA Symp 2000;200-04.
  17. Brossette SE, Sprague AP, Hardin JM, Waites KB, Jones WT, Moser SA. Association rules and data mining in hospital infection control and public health surveillance. J Am Med Inform Assoc 1998;5,:373-81.
  18. Prather JC, Lobach DF, Goodwin LK, Hales LK, Hage ML, Hammond WE. Medical data mining: Knowledge discovery in a clinical data warehouse. Proc AMIA Symp 1997;101-05.
  19. John Hayward. Mining Oncology Data: Knowledge Discovery in Clinical Performance of Cancer Patients. A Thesis submitted to Worcester Polytechnical Institute, Aug. 2006, MA 01609, United States.
  20. Hosking JR, Pednault EP, Sudan M. Statistical perspective on data mining. J Future Generaltion Computer System. 1997;13(3):117-34.
  21. Keim DA, Mansmann F, Schneidewind J, Ziegler H. Challenges in visual data analysis.
  22. Information Visualization. DOI:10.1109/IV.2006.31, 10th Int Conference 2006;9-16.
  23. Mannila H. Data mining: machine learning, statistics and databases. Paper presented at:8th J Int Conference on Scientific and Statistical Database Systems 1996.
  24. Bohacik J, Darryl ND. Estimation of cardiovascular patient risk with a Bayesian network. J Transcom 2011;27:129-32.