Chapter 7: Machine Learning: Developing Profiles


7.1 What Is Machine Learning?

As we have seen, there are a variety of AI technologies for combating crime. One of the most promising of them for the investigative data miner is machine-learning algorithms. These software programs can be used to develop profiles of perpetrators through a combination of decision trees and conditional IF/THEN rules. Unlike neural networks, which are at times difficult to decipher, using machine-learning and statistical algorithms like Classifier version 5 (C5.0), chi-square automatic interaction detection (CHAID), or classification and regression trees (CART), conditional constructs of criminal attributes and features can be extracted from large databases. For more information about the machine-learning technology go to http://www.mlnet.org.

The outputs from these and other machine-learning algorithms are highly descriptive in their classification of a desired solution, such as the conditions leading to criminal acts like fraud, or the attributes and features of criminals. For example, using one of these types of machine-learning algorithms, an analyst can generate a set of rules, such as the following one for profiling a potential smuggler at a point-of-entry border crossing:

       IF Vehicle Make is CHEVROLET,       AND Year of Vehicle is 1998,       AND No Insurance Listed for Vehicle,       AND Lien Holder is Owned,       THEN there is       06.34% chance that Alert is Low       18.32% chance that Alert is Medium       75.33% chance that Alert is High 

The rules are generated from an analysis of thousands of observations leading to a list of conditions (IF/AND) and a prediction (THEN) with a probability value associated to it (75.33%). The rules from these types of data mining analysis may be viewed as a ratio of conditions, which when combined lead to a predicted outcome with an associated probability. For investigative data mining, these types of machine-learning algorithms can lead to a wide range of criminal analyses and applications.




Investigative Data Mining for Security and Criminal Detection
Investigative Data Mining for Security and Criminal Detection
ISBN: 0750676132
EAN: 2147483647
Year: 2005
Pages: 232
Authors: Jesus Mena

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net