Question about Decision Trees and Neural Networks


  • I have some accounting data, with some transaction attributes and amounts.
    I'm using Decision Trees to try and predict the next month's amount for certain combinations of attributes.

    I've tried two different structures for the model:

    A: one with 9 discrete text input attributes.
    B: And another with the same 9 attributes + a avarage Amount for all combinations of the nine attribute for every transaction.

    When i've processed them and look in the dependency network, it says that the strongest link for the structure A is attribute "1".
    And for the second its the avarage-Amount attribute.
    Okey, that seems fine, but the second strongest link in structure B is attribute "2".

    Shouldn't it be attribute 1 like in structure A?

    Second question, if I run the same data in a Neural Network model, the prediction becomes much worst then the decision tree.
    I get many predictions that are negative values even though all training data contains positiv values.
    The StDev becomes the same for every row also..
    What am I doing wrong with that one. I have alot of transactions and a read somewhere that a Neural Network should work better than a decision tree in a case similar to mine.
    The score in the "Lift chart" for the Neural Network model becomes 0,00 and for Decision Trees with the same data I get around 110.
    terça-feira, 9 de maio de 2006 14:05


  • You might have a modeling issue here. If the individual attributes you have are transaction amounts, you should model them as continuous rather than discrete.
    quarta-feira, 10 de maio de 2006 06:28