The Algorithm That Catches Serial Killers

[douglas9]

“I wonder if we could teach a computer to spot serial killers in data,” Thomas Hargrove thought as he parsed the FBI’s annual homicide reports. The retired news reporter would soon answer his own question. He created an algorithm that, in his words, “can identify serial killings—and does.”

In The Dewey Decimal System of Death, a new film from FreeThink, Hargrove explains how “the real world is following a rather simple mathematical formula, and it’s that way with murder.”

The numbers are startling. According to Hargrove, since 1980, there have been at least 220,000 unsolved murders in the United States. Of those murders, an estimated 2,000 are the work of serial killers. Many of these cases are not ultimately reported to the Justice Department by municipal police departments; Hargrove has assiduously obtained the data himself. His Murder Accountability Project is now the largest archive of murders in America, with 27,00 more cases than appear in FBI records.

https://www.theatlantic.com/video/index/546893/serial-killer-algorithm/

[Loki Liesmith]

1. Not enough data

Too few examples of serial killer relative to population. False positives will be off charts.

[Jim__]

2. I don't think the number of serial killers relative to the population is the critical statistic.

The algorithm, as Hargrove describes it, doesn't actually identify the serial killer, but rather killings committed by a serial killer.

The first estimate of the critical statistic may be the proportion of homicides committed by serial killers. Using the data that he gives us, namely 2,000 by serial killer according to his algorithm relative to 220,000 total unsolved homicides - the algorithm is giving us approximately 0.9% of the unsolved homicides are by serial killer.

Police should have statistics that estimate the proportion of killings that are committed by serial killers - and you could probably add in some factor to account for the time the killings remain unsolved. That should give us an estimate of the false positives.