Machine learning has recently emerged as a promising alternative to programming for the
implementation of solutions to many real-world problems on computers. The learning approach offers
many advantages over the programming approach, particularly in contexts where information is
uncertain or incomplete. Indeed, programming presupposes that all of the knowledge necessary to the
solution of a given problem be available a priori. This assumption is valid in many well-defined,
closed domains, such as mathematics or accounting. However, there are many situations in which the
very nature of the knowledge (at least the representation thereof) is incomplete and uncertain, such as
medical diagnosis, pattern recognition, geological prospection, and many decision-making applications such as
loan underwriting, and mine-rock discrimination (Giraud-Carrier,C. and Martinez, T.).
The nature of these applications necessitates the use of prior knowledge, where such knowledge is presented as examples.
The use of prior knowledge is one of the fundamental approaches in machine learning.
In order to select an effective machine learning model, we tested a number
of algorithms that use prior knowledge for classification. Such algorithms are:
• Neural Networks:
Neural networks are inspired in
part by observations in biological
learning systems. They are built on densely interconnected set of simple units
(neurons), and are a robust approach
for certain problems such as handwritten examples recognition, and image
recognition. Neural networks are appropriate where:
--cases are represented by multiple attributes of varying importance
--training examples may contain errors
--long training times are acceptable
--the ability of humans to understand the learned target function is not important, or not easily
communicated to humans.
• Decision Trees:
Decision trees present one of the most widely used and practical methods for inductive inference. The advantages with
using decision trees are as follows:
--they are robust to noisy data.
--they are capable of learning disjunctive expressions.
--the training data may contain errors or inaccuracies.
--the training data may contain missing attribute values.