How can the data be visualized to support interactive decision tree construction?

Data MiningDatabaseData Structure

Perception-based classification (PBC) is an interactive method based on multidimensional visualization methods and enable the user to incorporate background knowledge about the data when constructing a decision tree.

By optically interacting with the data, the user is likely to produce a deeper learning of the data. The resulting trees likely to be smaller than those construct utilizing traditional decision tree induction techniques and therefore are simpler to interpret, while achieving about the similar accuracy.

PBC need a pixel-oriented method to consider multidimensional data with its class label data. The circle segments method is adapted, which maps d-dimensional information objects to a circle that is divided into d segments, each defining one attribute.

An attribute value of a data element is mapped to one colored pixel, introspective the object’s class label. This mapping is completed for every attribute–value pair of every data object. Sorting is completed for each attribute to decide the arrangement series within a segment.

For instance, attribute values within a given segment can be organized so as to shows homogeneous (concerning class label) regions within the similar attribute value. The amount of training information that can be anticipated at one time is approximately decided by the product of the multiple attributes and the multiple data objects.

The PBC system shows a split screen, including a data interaction window and a Knowledge Interaction window. The data interaction window shows the circle segments of the data under checking, while the Knowledge Interaction window shows the decision tree built so far. The entire training set is visualized in the data interaction window, while the Knowledge Interaction window shows a null decision tree.

Traditional decision tree algorithms enable only binary splits for statistical attributes. PBC enables the user to define multiple split-points, resulting in several branches to be grown from an individual tree node.

A tree is interactively built as follows. The user anticipate the multidimensional data in the data Interaction window and choose a splitting attribute and multiple split-points. The modern decision tree in the Knowledge Interaction window is expanded.

The user choose a node of the decision tree. The user can create a class label to the node (which create the node a leaf) or request the visualization of the training information equivalent to the node. This leads to a new visualization of each attribute except the ones used for splitting element on the similar path from the root. The interactive procedure continues until a class has been created to every leaf of the decision tree.

Updated on 18-Feb-2022 07:20:57