Data Mining - Decision Tree Classifier and Pattern Mining.

Develop a single [login to view URL] file to perform following 3 tasks. Each task is invoked using a different command line command.

Task 1: Build a decision tree classifier for an input dataset in the form of redwine.csv. Assume all attributes except the class are numerical, and the last attribute is the class. The output should be written to a file called [login to view URL] describing the accuracy of the classifier when tested on the input dataset and also the parameters used to build the classifier. [login to view URL] will perform this task when given these 2 command line arguments: 1, [login to view URL]; here [login to view URL] is the name of an input dataset.

Task 2: Find top 10 patterns from an input dataset D similar in form to the itemized data file produced in previous project (refer to [login to view URL]), except that the class values may be categorical. Assume the values of the last column are the classes. For convenience, here we refer to one class as P and the other as N. If the input file uses other class names, treat the lexically smaller as P and the other as N. (Assume there are just two classes.) The top 10 patterns should contain the top 5 among the single item patterns, plus the top 5 among all non-singleton patterns. The expectation is that you can find the correct top 5 single-item patterns, but you might not be able to find the true top 5 among all non-singleton patterns. The ranking is determined using the function GR(X) * sup(X,D), where X is a pattern and GR(X) is defined as (sup(X,P)+1)/(sup(X,N)+1). The top 10 patterns must be written to an output file called top10EPs.csv. For each of top 10 patterns X={x1,x2,…,xk}, there should be a line in this file containing the following:

GR(X) * sup(X,D), GR(X), sup(X,D), x1,x2,…,xk

[login to view URL] will perform this task when given these 2 command line arguments: 2, [login to view URL]; here [login to view URL] is the name of an input dataset.

Task 3: Given an input dataset D3 similar in form to [login to view URL] and a pattern X, build a decision tree classifier on mds(X,D3). Recall that mds(X,D) is the subset of instances in D satisfying all single-attribute conditions in X. The output is written to a file called [login to view URL] describing the accuracy of the classifier when tested on mds([login to view URL]) as a test dataset, the number of instances in mds([login to view URL]), and also the parameters used to build the classifier. [login to view URL] should perform this task when given these 3 commandline arguments: 3, [login to view URL], X

Here [login to view URL] is the name of an input dataset and X is a pattern. When specifying patterns, we will refer to the attributes as A1,…,An, in the left to right order as they appear in the input data. Any pattern will be described in the following form:

Ai1 CMP Val1, Ai2 CMP Val2, …, Ai3 CMP Val3

Here CMP is either < or >, Ai1, Ai2, …, Ai3 are attributes, and Val1, …, and Val3 are numerical values. A pattern can contain one or more attributes. An example pattern is A3<5, A6>4.

All outputs are written to the folder where the jar file is located.

Correctness, efficiency, and quality of the findings (for Task 2) will be important factors for marking and evaluation.

You should use Weka for classification and you can use any frequent pattern mining programs written in Java in your program. You cannot use emerging/contrast pattern mining programs written by others; of course you can implement your own.

Include a [login to view URL] file indicate which frequent pattern mining program you are using (indicate developed by whom and the source URL).

All the above information is included in [login to view URL]

データマイニング 情報処理 Data Science Data Analytics Java

プロジェクトID: #24443923

プロジェクトについて