Neural network is a data mining tool that derived from the idea of human brain neural capability. In order to use it, it has to go through a learning phase where data are input into a predefined set of computational mathematics algorithm. After the learning phase, user can uses the neural network to make prediction based given any input.
In contrast, decision tree requires the user to know the input and the output of the prediction. It is easier to understand as it shaped with logical and sequential based of prediction.
Both of this technique is important for data mining but this time, I will explore which tool is better in terms of making a prediction.
First of all, neural network has it own shortcoming such as incomprehensible for most people and validating the prediction is hard. User will have no idea how the neural network come out with a certain result. However, this technique is basically categorising different data into sensible information and presented it to the user. By suggesting categorising different data, it has step into the boundary of decision tree model. This is because decision tree is also about categorising data into different categories and trying to making sense out of it. This can be seen that both techniques is actually the same. It is different in term of developing and implementing it but the general idea of how to create information for the user is the same.
My point is that regardless how many tools and techniques out there, data mining tools and techniques has not changed since it was introduced centuries ago where people doing it without computer. Essentially, data mining is always about categorising data into different categories to enable the decision maker to make the judgment. Although different tools and techniques use different algorithms, they are still supporting the same idea of categorising data. In short, there is no one superior tool or technique when it comes to data mining. It is more on reinforcing a precision of a prediction by applying different algorithm.
Subscribe to:
Post Comments (Atom)
2 comments:
Isn't improved accuracy (and not to mention speed in processing compared to manual analyses) a good thing?
Cheers,
Rob
Regarding your first two paragraphs: Your use of the phrase "in contrast" suggests that construction of neural networks does not require training examples which include inputs and known outputs (targets). Generally (except with clustering neural networks), artificial neural networks will require the same sort of data that decision tree-induction does.
Post a Comment