Data Mining by Mehmed Kantardzic (good book recommendations TXT) 📕

system combines data from all available systems. This includes extracting knowledge from the unstructured clinical records. The REMIND system combines all available sources of data present, then using redundancy in data the most likely state of the patient is found. For example to determine a patient is diabetic one may use any of the following pieces of data: billing code 250.xx for diabetes, a free text dictation identifying a diabetic diagnosis, a blood sugar value >300, a treatment of insulin or oral antidiabetic, or a common diabetic complication. The likelihood of the patient having diabetes increases as more relevant information is found. The REMIND system uses extracted information from all possible data sources, combined in a Bayesian network. The various outputs of the network are used along with temporal information to find the most probable sequence of states with a predefined disease progression modeled as a Markov model. The probabilities and structure of the Bayesian network are provided as domain knowledge provided beforehand by experts and tunable per deployment. The domain knowledge in the REMIND system is fairly simple as stated by the author of the system. Additionally, by using a large amount of redundancy the system performs well for a variety of probability settings and temporal settings for the disease progression. However, before a wide distribution of the REMIND system a careful tuning of all the parameters must take place.

One example deployment was to the South Carolina Heart Center where the goal was to identify among 61,027 patients those at risk of Sudden Cardiac Death (SCD). Patients who have previously suffered a myocardial infarction (MI; heart attack) are at the highest risk of SCD. In 1997 a study was performed on the efficacy of implantable cardioverter defribillators (ICDs). It was found that patients, with a prior MI and low ventricular function, had their 20-month mortality rate drop from 19.8% to 14.2%. This implantation is now a standard recommendation. Previous to the REMIND system one had two options to find who would require the implantation of an ICD. The first option is to manually review the records of all patients to identify those who were eligible for an ICD. This would be extremely time-consuming considering the large number of records. The other approach would be to evaluate the need for an ICD during regular checkups. However, not all patients come in for regular checkups and there would be a high chance that not every patient would be carefully considered for the need of an ICD. The REMIND system was given access to billing and demographics databases and transcribed free text including histories, physical reports, physician progress notes, and lab reports. From these data the REMIND system processed all records on a single laptop in 5 h and found 383 patients who qualified for an ICD.

To check the validity of the 383 found patients, 383 randomly chosen patients were mixed with the 383 found previously. Then 150 patients were chosen from the 766 patient samples. An electrophysiologist manually reviewed the 150 patients being blinded to the selection made by the REMIND system. The REMIND system concurred with the manual analysis in 94% (141/150) of the patients. The sensitivity was 99% (69/70) and the specificity was 90% (72/80). Thus it was shown that the REMIND system could fairly accurately identify at-risk patients in a large database. An expert was required to verify the results of the system. Additionally, all of the patients found would be reviewed by a physician before implantation would occur.

From the previous cases we see that a great deal of time was required from experts to prepare data for mining, and careful analysis of a model application needed to take place after deployment. Although applied data-mining techniques (neural and Bayesian networks) will be explained in the following chapters, the emphasis of these stories is on complexity of a data-mining process, and especially deployment phase, in real-world applications. The system developed for Banmedica was measured after analysis in terms of fraudulent cases found and the amount of money saved. If these numbers were not in favor of the system, then it would have been rolled back. In the case of the REMIND system, the results of the system wide search had to be manually analyzed for accuracy. It was not enough that the rules were good, but the actual patients found needed to be reviewed.

4.10 REVIEW QUESTIONS AND PROBLEMS

1. Explain the differences between the basic types of inferences: induction, deduction, and transduction.

2. Why do we use the observational approach in most data-mining tasks?

3. Discuss situations in which we would use the interpolated functions given in Figure 4.3b,c,d as “the best” data-mining model.

4. Which of the functions have linear parameters and which have nonlinear? Explain why.

(a) y = a x5 + b

(b) y = a/x

(c) y = a ex

(d) y = ea x

5. Explain the difference between interpolation of loss function for classification problems and for regression problems.

6. Is it possible that empirical risk becomes higher than expected risk? Explain.

7. Why is it so difficult to estimate the VC dimension for real-world data-mining applications?

8. What will be the practical benefit of determining the VC dimension in real-world data-mining applications?

9. Classify the common learning tasks explained in Section 4.4 as supervised or unsupervised learning tasks. Explain your classification.

10. Analyze the differences between validation and verification of inductive-based models.

11. In which situations would you recommend the leave-one-out method for validation of data-mining results?

12. Develop a program for generating “fake” data sets using the bootstrap method.

13. Develop a program for plotting an ROC curve based on a table of FAR–FRR results.

14. Develop an algorithm for computing the area below the ROC curve (which is a very important parameter in the evaluation of inductive-learning results for classification problems).

15. The testing data set (inputs: A, B, and C, output: Class) is given together with testing results of the classification (predicted output). Find and plot two points on the ROC curve for the threshold values of 0.5 and 0.8.

16. Machine-learning techniques differ from statistical techniques