Rapidminer model - open to bidding

Hi,

I have a dataset (included in the attachment of this message) containing 5 keywords (a-e) with the respective google positions (1-20) and 7 features that influence the google position:

name (data type)

--------------------

PR (0-10)

BM25 (%)

Number of links (int)

Percentage match (%)

Match domain (boolean)

IP (boolean)

Number of other (int)

Based on the 7 features the model should predict the google position on unseen data as precise as possible. I'm looking for some one that can build this in a Rapidminer model.

The model should do the following:

- import the data from CSV

- normalize the data (how to deal with the booleans?)

- split the data into a test, training and validation set.

- select which features to use for training using a "greedy approach" (first train the model on each individual feature, then start with the strongest feature and add one feature a time in order to see the best combination of features). Other suggestions to calculate the optimal selection of features (such as calculating the information gain) are welcome as well.

- train the model using 4 different techniques: SVM, Decision Trees, Logistic Regression and Lineair Regression

- analysis of which of the 4 techniques delivers the best performance.

Please include in your application a short indication of how you would deal with the booleans, what your approach for selecting the optimal combination of features is and your previous experiences with rapidminer.

Looking forward to your application!

Kind Regards,

Dirk

Data Mining Data Processing Java Research Statistics

Project ID: #7879086

About the project