AKD (Adaptive Knowledge Discovery)

Overview
This project is to design, develop,and test a creative and stable software tool, which could finally enable the knowledge discovery as soon as the industry case data is given. The components in format of DLL are being developed using Visual C++, including normalization, discretization (Chi2), dimensionality reduction (based on statistics), neural network modeling, rule extraction from decision tree and clustering, AMFM rule tuning (Adaptive Mamdani Fuzzy Model), as well as rule base pruning.
Installation and Operation manual are available for downloading.

Team
Dr. Samuel Huang (Principal Investigator) - Sam
Dr. Jun Shi(Project Manager & Principal Programming) - John
Ranganath Kothamasu (Algorithms and Debugging) - Ranga
Saurabh Dwivedi (Testing) - Saurabh

Instructions

(1) Start Program

• Double-click AKD.exe
• Get window as following:
  
• The window includes four parts, indicated as the following drawing:
  
• The top part helps the user select one of the demo cases. The user can open other case data after clicking the the Wizard button. End the program by clicking on the Close button.
• Left part is a list of big buttons; each of them is related to one function of the program.
• Middle part contains all file names and parameters values that will be used by various functions.
• Right part is a group of small buttons. They will pop up dialogs to show data grid, and compile data such as rules and terms.

(2) Run Functions

• Normalization:
  Set names of the initial data file and the normalized data file.
  Click the Normalization button.
  The program reads the initial data file and generates the normalized data file.
  
• Discretization:
  Set names of the initial data file, the discretized data file and the interval list data file.
  Click the Discretization button.
  The program reads the initial data file, and generates the discretized data file and the interval list data file.
  
• Rule Extraction with Decision Tree:
  Set the name of the rule file.
  Click the DT Extraction button.
  The program reads the discretized data file, and generates rules.
  
• Rule Extraction with Clustering:
  Set names of the rule file and the term file.
  Click the Clustering Extraction button.
  The program reads the original data file, and generates rules and the term file.
  

(3) Wizard

• Click the Wizard button.
  
  The rule extraction dialog pops up.
  
• The dialog opens the original file selected in the Demo Cases radio buttons by default.
• The user clicks the Open button to get another original data file.
  
• The user can set the flow chart of the program: normalization is needed or not; discretization is needed or not; rules are extracted using decision or clustering; and AMFM is used to tune the rules or not.
• Click the Extract rules button to run the process.
• Now the View rules button and the Save rules button are enabled.
• Click the View rules button, the Rule explorer dialog will pop up, which will be described in the rest of the manual.
  
• Click the Save rule button to save rules into a text file, which can re-used in the Rule explorer dialog.
  
• End the dialog and return to the main window by clicking on the Close button.

(4) Show Data Grids

• Click the Show Data button on the right hand side of the original file name text box;
  
  The Initial data dialog pops up, as shown in the following figure.
  
• The same operations apply for displaying the normalized data.
  
• The same operations apply for displaying the discretized data.
   
• Click the Show lists button and the interval list data shown in format of grid as in the following figure.
  
• The same operations apply for displaying the rule data extracted by decision tree.
  
  The dialog is shown as in the following figure.  Each row of the grid data is a rule. The values show the position of the parameter's value in the interval list. "~" means the parameter does not appear in the certain rule. So the first line in the dialog is not a valid rule. The second line means that "If sepal length=0 and sepal width=1 Then iris class=0 ". 
  
• Click the Show Data button on the right hand side of the rule file name text box for rule extraction with clustering.
   
  The rule data dialog pops up, as shown in the following figure. Each row of the grid data is a rule. The values show the position of the parameter's term array. The term array stands for the logistic terms the certain parameter has. For example, the sepal length may have three terms ["middle", "short", "long"]: "middle" in position 0 of the array; "short" in 1; and "long" in 2, Thus, the first line means that "If all of sepal length, sepal width, petal length and petal width are the first terms in their term arrays, Then iris class=0 ", so on. 
  
• Click the Show terms button and the terms data shown in format of grid as in the following figure.
  
• For each data grid dialog, click on the Close button to return back to main window.

The rules and terms have a better facilities to operate, which will be address in the rest of the manual.

(5) Compile Terms

• Click the Define terms button on the right hand side of the interval list file name text box, to display and compile the linguistic terms generated by discretization.
  
  The term explorer dialog pops up.
  
• The attribute Combo box let the user select the parameter to work on.
  
• The major part of the dialog is a two column list view: the left column lists the default names for items in the interval list. The user can assign a name for it on the right hand side.
  
• Click the Update button to accept the changes in the list view part for current parameter. Or the changes will not affect the results after the user switch to other parameter or close the dialog.
• Click the Retrieve button to recover the values changed in the list view.
• Click the Cancel button to close the dialog and return back to main window without saving data.
• Click the Save & Return button to save the changes and return back to main window.
• Save operations applies on the terms generated in clustering rule extraction.
  
  
  

(6) Compile Rules

• Click the Show rule button to display and compile rules extracted through decision tree.
  
  The Rule explorer dialog pops up as shown in the following figure.
  
• The Rule explorer opens the file selected in the main window by default.
• Click on the Open button to open another rule data file.
  
• The top side list box shows all rules in  the IF-THEN format. Click on each item, IF part and THEN part of the associated rule will be shown in the bottom side.
• The user can change the values in the IF part list view and THEN part list view.
• Click the Update button to accept the changes; the associated item in the top side list box will be changed.
• Click the Add button to append a new rule in database; then the user can update the new rule.
• Click the Delete button to remove a rule form the database.
• Click the Save button to save all changes into the rule data file.
• Click the Close button to return back to the main window.
• Save operations apply on the rules generated through the clustering method and the wizard facility.