onfidence for the inference. For instance, if a novel data point is

the upper-left subspace shown in Figure 3.37(b), this novel data

hen labelled by or predicted as a new member of the cross class

almost 100% confidence. This is because all data points in this

belong to the cross class. Of course, if a novel data point is

n the upper subspace generated using the partitioning rule y = 0 in

37(a), the confidence of labelling this data point to the cross class

ower because this subspace has been far less pure for the cross

ose a partitioning strategy for a data space has been derived, how

artitioning space be expressed or visualised as a decision-making

r interpreting every decision which has been made using the

The plots in Figure 3.37 are certainly not very efficient to use. In

explore human intelligence of a model, a tree-like decision-

tructure has been adopted by the inductive learning algorithms.

ch a tree, how a decision is made can be clearly visualised in the

by step. Importantly, a decision-making process is sequentially

d. For instance, such a tree can show which variable is employed

nd which variable is employed in the next steps to explain why

a decision is made. Finally, such a tree can show how confident

a decision which has been made. In short words, an inductive

model can help deliver a human-intelligence-alike model.

(a) (b)

The decision-making trees for the partitioning rules derived for the data shown

37. A diamond represents a decision-making process using a partitioning rule.

presents a made decision. (a) The tree model constructed for the data shown in

(a). (b) The tree model constructed for the data shown in Figure 3.37 (b).