6.13 False Positives

In the movie Minority Report the mythical Precrime Department is able to foretell homicides before they occur and arrest perpetrators before they commit crimes through a process of previsualization. However, something goes wrong and the leading character (a precrime officer) becomes the wrongly accused suspect: he becomes a false positive. A false positive is a false alarm or false detection and is part of making predictions and classification. Models are not perfect and neither are the neural networks that they are constructed from. Care must be taken to ensure that rigorous testing takes place during the construction of predictive models from neural networks to ensure that false positives are minimized and innocent individuals are not wrongly accused of crimes.

In this chapter we have described several case studies demonstrating the ability of neural networks to predict at-risk police officers, identify potential smugglers, discover the signature of arson, and associate modus operandi to a network of criminals. While this is encouraging, many more applications await where neural network and other technologies can and will be used to intercept terrorists and discover patterns of criminal activity. Care, however, must be taken to ensure that the privacy and legal rights of individuals are not violated and that models are within the law of the land. The technology should be used to ensure that false positives are reduced, rather than increased, and that law-abiding citizens are protected from perpetrators.

As more and more information about individuals becomes available online, the need for robust and reliable means of verifying identity also increases. One of the technologies that is receiving considerable attention in the aftermath of 9/11 is biometrics—the use of the physiological traits or characteristics of individuals as a basis for recognizing and confirming a person's true identity. Neural networks are the core technology of these pattern-recognition engines used in fingerprint and facial recognition, retinal scanning, voice-prints, and even the monitoring of keystroke typing patterns or handwriting stroke patterns.

Man has from the beginning of time survived by recognizing the features and traits of his enemies, using the best available weapons, which today include neural networks. In addition to individual identification, ubiquitous Internet connections require that organizations whose internal computer systems connect to the Web also be prepared to prevent malicious or unwanted intrusion into internal systems or open public sites. Neural networks can be used to learn patterns of acceptable packet contents and arrival rates and, thus, serve as a dynamic security firewall.