Question

Q 1.Explain steps to be performed in decision tree classification? Alsoexplain the

conditions where it needs to stop.Marks5

Q 2.Briefly explain the properties of k-means algorithmMarks5

Q 3.How a Self – Organizing Map (SOM) could be used to analyzedata in a data

warehouse? Explain briefly.Marks5

Answer 1

Q1.  

A decision tree classifier is built in two phases:

A growth phase

A prune phase

The tree growing phase is an iterative process whichinvolvessplitting the data into progressively smaller subsets.Eachiteration considers the data in only one node. The firstiterationconsiders the root node that contains all the data.Subsequentiterations work on derivative nodes that will containsubsets ofthe data.

The process of pruning the initial tree (built in growthphase)consists of removing small, deep nodes of the tree resultingfromnoise contained in the training data, thus reducing the riskofoverfitting, and resulting in a more accurate classificationofunknown data.

Tree-building algorithms usually have several stoppingrules.These rules are usually based on several factors includingmaximumtree depth, minimum number of elements in a node consideredforsplitting, or its near equivalent, the minimum number ofelementsthat must be in a new node. One is to stop splittingbeforethe nodes are pure.

Q2.  

The k-means algorithm has the following important properties

(a) In its basic form, it works only on numeric values.

(b) It uses the Euclidean metric and hence the centroid of Cisthe mean of the points in C.

(c) It is efficient in processing large data sets.Thecomputational complexity of the algorithm is O(nkt), where n isthetotal number of objects, k is the number of clusters and t isthenumber of iterations. In clustering large data sets, thek-meansalgorithm is much faster than the hierarchicalclusteringalgorithms, whose general computational complexity isO()

(d) It terminates at a local optimum.

Q3.

The SOM technique creates a two-dimensional mapfromn-dimensional input data. This map resembles a landscape inwhichit is possible to identify borders that define differentclusters..The self-organizing map describes a mapping from ahigherdimensional input space to a lower dimensional map space.SinceSelf-Organizing Maps (SOM) provide compact representation ofthedata distribution, efficient process monitoring can be performedinthe two-dimensional projection of the process variables.

SOM has the ability to do drill-down processing. Indrill-downprocessing data is arranged in layers so that access andanalysisof one layer can lead to another layer

SOM quickly relates documents. Once the analyst has examinedtheSOM and if you want to look at the document then direct accessofthe document is allowed.