The research project will materialize everything we will study during the course. You will feel the application of the concepts we are studying. I. Data Analysis Project. Identify the problem(s) to be solved or opportunities to be realized by mining the selected data set. Consider the following data preparation questions and explain your answers. When appropriate cite resources that support your answer. Explain how the answers, and data preparation, differed when you chose a different data mining method. Should instances with missing values be deleted? Should missing values be specially coded and then retained in the data set? Should numeric values be assigned predetermined ranges or left for the algorithm to split? Should categorical variables be grouped or coded to reflect a hierarchy? To explore the problem or opportunity, use two or more of the following data mining methods covered by this course: regression: linear regression, discriminant analysis or logistic regression, decision trees, neural networks, hierarchical or k-means clustering, association rules, time series, genetic algorithms. Describe the algorithms chosen, and indicate why you chose them. Exploring a method of interest is a satisfactory reason for this course paper. Explain how and why you used specific pruning parameters or other adjustments to create a sparser model. Compare the alternative solutions using methods found in comparative studies in the literature. For example, see “Data mining for network intrusion detection: A comparison of alternative methods” Dan Zhu, G Premkumar, Xiaoning Zhang, Chao-Hsien Chu. Decision Sciences.Atlanta: Fall 2001.Vol.32. http://www.findarticles.com/p/articles/mi_qa3713/is_200110/ai_n8954240 Report the results of the accuracy measures available with the software. If the software used does not have built-in accuracy reporting then manually test the model’s accuracy on a small hold-out test sample of the data. The hold-out method creates separate training and test sets. This is particularly useful when testing the model on data from a later time period. Create a table showing the number of cases correctly identified, Type I, and Type II errors. In addition, a ROC curve is appropriate with discriminant analysis and logistic regression. For these methods, changing the parameters for the line separating the classes, changes the percentages of Type I and Type II errors. Medical practitioners like ROC curves because they show the tradeoff between false positives and false negatives. Which data mining method(s) seem superior for the chosen data set? Did the method that performed best in your study also dominate in similar comparative studies? Compare the results or recommendations that would result from the use of the different methods. Based on your analysis, justify a conclusion or recommendation. Cite the relevant literature using APA formatting described at http://www.umuc.edu/library/libhow/citeright_tutorial_apa_articles.cfm Any publication listed in the references should be cited in the paper. Organize the paper into the sections of a formal research paper: Introduction, Methods, Results, etc., Use the resources provided in the Effective Writting Center: http://www.umuc.edu/writingcenter/ This graded exercise represents your point-in-time progress towards the completion of the course project. At this point in the project, you should have the following items accounted for with a shell of a Word document representing your final project: Title Page Abstract Introduction Background Information/Domain-Specific Discussion Dataset Selection Method Selection (at least two) Data Preparation/Data Preprocessing Method Analysis (each) Method Performance (each) Comparison to Similar Studies in Literature Conclusion References This is a general guide. Use the other course project resources to get an idea of what sections you require in order to meet the minimum requirements of this project. We do not expect these items to be completed, i.e. not in the their final form. However, you should have these items blocked out, researched, brainstormed, and structured. For example, your abstract may be as simple as a few notes on your paragraph structure accompanied by the keywords you expect to use. Your introduction may be blocked similarly with notes about key facts you wish to highlight or research you mean to reference along with your thesis statement. Your dataset must be included in this deliverable in the form you will be analyzing it in. This means your dataset selection/preparation/preprocessing stages will include not only the original data but any modifications that your method selections will require. You may have to perform different modifications on the original data to get your different method selections to work. This is the section where you make note and execute this. Each model should be accompanied by justification for why that model was selected for your analysis and why. At this stage, you may or may not have actually started to execute your selected methods against the dataset. It is not expected so you do not need to include anything in the model performance section unless you already have it done. The comparison to similar studies in published literature is vitally important. If you have not already, you should find one or more companion studies that have attempted to solve the same problem that you are solving (but in a different way). You are expected in your final project to compare the performance (results) of their study against yours. Once compared, you will draw conclusions as to why your model performed better/worse than theirs and what you may do to improve. As with the introduction, the conclusion should be blocked out with ideas or hypotheses related to your expectations. The more you have the better as this is a graded ‘check-for-progress’. Even if you are lite on content, the purpose of this deliverable is to help get you to where you need to be prior to the start of the final weeks of class.