Bad cohen's kappa


I am a fairly new KNIME user, but I would like to learn more about this program.

To train ourselves I am using a database(uploaded here below) that collects information about different types of customer. The Q that stands before each category stands for the ordered quantity of a certain product. The R that stands before each category stands for number of days after their last purchase. The C that stands before each category tells us , wether or not a customer has bought the product during the observation product.

Our job is to predict the correct outcome of the value of C6 and C9 , so basically wether or not a customer will buy the C6 product or the C9 product.

So my main question is how to increase the performance of my model to have a better outcome and from what I've heard a better cohen's kappa. (Currently I have a cohen's kappa of 0.58 for the C6 variable and 0.15 for the C9 variable)

One of the issues I am currently facing , is the problem of the variables. I would like to detect which variables are unneccessary to the model so that I could improve my predictions.

Another issues I currently have is the combining of different models, how can I best combine different models?

Thank you in advance for any help,


Hi KNIME-ator,

Maybe you could use one of the Feature Selection Nodes available in the Node Repository under Analytics/Feature Selection. Additionally, have you checked the video on Data Preparation from our E-Learning Course that describes several common data preprocessing steps? Another video on Analytics talks about different metrics to evaluate performance of the model.

Hope it helps.

Best, Daria