London's Crime Hot Spots Predicted Using Mobile Phone Data 64
KentuckyFC (1144503) writes A growing number of police forces around the world are using data on past crimes to predict the likelihood of crimes in the future. These predictions can be made more accurate by combining crime data with local demographic data about the local population. However, this data is time consuming and expensive to collect and so only updated rarely. Now a team of data experts have shown how combing crime data with data collected from mobile phones can make the prediction of future crimes even more accurate. The team used an anonymised dataset of O2 mobile phone users in the London metropolitan area during December 2012 and January 2013. They then used a small portion of the data to train a machine learning algorithm to find correlations between this and local crime statistics in the same period. Finally, they used the trained algorithm to predict future crime rates in the same areas. Without the mobile phone data, the predictions have an accuracy of 62 per cent. But the phone data increases this accuracy significantly to almost 70 per cent. What's more, the data is cheap to collect and can be gathered in more or less real time. Whether the general population would want their data used in this way is less clear but either way Minority Report-style policing is looking less far-fetched than when the film appeared in 2002.
Re: (Score:1)
This has nothing to do with caucasions.
Re: (Score:2)
"In other words, they found where the poor people live by looking at phone data."
No they found out that crime happens where the criminals are.
Groundbreaking.
Price of safety (Score:1)
Re:Price of safety (Score:5, Insightful)
Crime reduction is certainly a worthy reward, but as the article says, lots of people might not be too happy with having their information shared this way.
Especially considering that said "information sharing" leads to a mere 8% increase in accuracy.
Let's hope it is truly anonymous (which I doubt) and see how it goes.
Let's assume that it's not, and see how it's used nefariously. That's not cynicism, that's realism.
Re: (Score:2)
Re: (Score:3)
Yes, it is. Your "privacy" is not worth a human life. And no, you don't get to have any say in the matter.
Sayeth the Anonymous Coward.
Why not include your name, address, and contact info on every post? after all, your "privacy" is not worth the chance that you might someday take a human life, right jackass?
Re: (Score:2)
Meh. I see that argument used frequently on /.
... and you've done nothing to prove the aforementioned argument wrong. Good job.
Re: (Score:2)
Your "privacy" is not worth a human life.
Right. Because every crime takes a human life.
Re: (Score:1)
Everyone's privacy must be ignored because doing so might save a life somewhere, sometime.
Re: Price of safety (Score:4, Interesting)
Re: (Score:1)
Re: (Score:2)
Typical statistics (Score:3)
Expounding on your statistics point as I agree that there is no significant increase in accuracy, notice the key phrase in the article.
The team used an anonymised dataset of O2 mobile phone users in the London metropolitan area during December 2012 and January 2013. They then used a small portion of the data to train a machine learning algorithm to find correlations between this and local crime statistics in the same period.
In other words, they took everything they gathered and pulled a subset that matched criteria that would back the claim that they could detect future crimes.
Computers can surely show what law enforcement already knows. E.G. That area is a known crime area. Computers don't make tea leaf reading possible, which is the claim that both Governments and Tech companies peddling so
Re: (Score:3)
While it's possible that they did in fact pull a biased sample, this methodology is what I was taught in academia as a legit way to test machine learning. If you have one sample set, first split it into two. Use one set, usually much smaller, to train the neural network. That data set, because it's tuned to find those specific correlations, obviously
Re: (Score:2)
They did not submit a smaller sample as academia would teach, they submitted a small set of "select" data (their words, not mine).
If you are only teaching a bias only a bias will be understood.
Re: (Score:2)
I read "select" as meaning specific fields (as in, select types of data), not deliberately selected subsets of data... got a quote that helps clarify things?
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Speaking of accuracy, it went from 62% to 70%. While indeed that is eight percentage points, it's a nearly 13% improvement.
Re: (Score:2)
Point being, my privacy is worth more than 13%.
Re: (Score:2)
62% to "nearly 70%"
That could easily be 66% and optimism.
Re: (Score:2)
Well, closer to 22%. While it's true that 8% of the predictions are more accurate, what is important is that ~22% of the predictions that used to be wrong are no longer. In much the same way as if it went to 100% accurate, you don't get to bitch about it being only a 38% increase in accuracy. You get to talk about whether it's worth the cost, and how we can get something only 62% as accurate without the cost.
phone app auto tracks, health, academics, behavior (Score:1)
New Dartmouth smartphone app reveals users' mental health, performance, behavior
Dartmouth researchers and their colleagues have built the first smartphone app that automatically reveals students' mental health, academic performance and behavioral trends. In other words, your smartphone knows your state of mind -- even if you don't -- and how that affects you.
The StudentLife app, which compares students' happiness, stress, depression and loneliness to their academic performance, also may be used in the gener
Re: (Score:2)
I don't buy it - an app that monitors every sensor, plus apparently monitoring abstract stuff like "stress level" somehow, 24/7?
Wouldn't that pretty much lock up and drain the battery of almost every phone on the market today? Hey, maybe that's how they determined stress level - using the accelerometer to determine how hard the student threw the phone against the wall when it froze up on them for the last time.
Sorry, but no. (Score:1)
"Let's hope" increasingly is not good enough. There is a point where just going forward without applying what we learned or even learning from what happened before in similar situations (your "hope" here) becomes criminal negligence. If we're not past that point for everybody yet, and apparently not past it for those who should be paying attention to this sort of thing, we're certainly past it for those who do pay attention to what happens in this space.
Re: (Score:2)
Let's hope it is truly anonymous
Some interesting data could still be collected. If the same phone repeatedly appears near the scene of a crime, one could deduce that crimes will occur in the future in its proximity.
From TFA:
Their analysis shows that some mobile phone data is more important than others. For example, the data relating to whether or not the phone owner was at home, was particularly strongly correlated with crime patterns.
Not so anonymous, IMO.
Here's the algorithm (Score:3, Insightful)
Re: (Score:2)
Which would just be a notification to the would be thieves that you have a phone worth stealing!
8pts? (Score:2)
fuck them. Almost 2/3 prediction from existing crime stats. Gee I know a lot of cops aren't the brightest but really? Thats not enough of a leg up?
Milk (Score:3)
The "machine learning algorithm" is a euphemism for three hairless teenagers floating in pools of milk.
Watch out for the spiders.
Doubtful (Score:4, Interesting)
62% to 70% isn't exactly groundbreaking for something that varies greatly. This increase looks suspiciously like selecting results for passing a statistical test instead of using a statistical test to verify the significance of a given result. Relevant xkcd: Significant [xkcd.com].
Also, there is no such thing as anonymised phone data.
Criminal's phones (Score:3)
Re: (Score:2)
Or as an alternative: If you track the location of cop's cell phones you can predict areas at higher risk for crimes, after they've been called in.
Percent. . .Percent. . . PERCENT! (Score:2)
Any article citing statistics is invalid when they don't understand the difference between percent and per cent. Getting 62 things right per US penny is a VERY cost effective system, probably regardless of what information we want to get right.
Unfortunately, all this says is that if we place our population under total surveillance with trackers, we can increase anticipation of crime by 8% (accuracy of 62 to ALMOST 70%). This says nothing about preventing those crimes or what type of crimes it prevents.
Re: (Score:2)
I expect it's protection against invasion of privacy is limited.
Re: (Score:1)
FYI: "The one-word percent is standard in American English. Percent is not absent from other varieties of English, but most publications still prefer the two-word per cent. The older forms per-cent, per cent. (per cent followed by a period), and the original per centum have mostly disappeared from the language (although the latter sometimes appears in legal writing).
"There is no difference betw
Re: (Score:2)
Well, I will consider myself schooled! Thank you for educating me. That has always been a huge annoyance of mine and many others I know, but I guess it actually does make more sense when considering the origin of the word. I am saddened that one of my huge pet peeves is apparently unjustified, but in time I will adjust.
On the other hand, I still love finding unnecessary quotes [unnecessaryquotes.com] in public!
I see you.... (Score:2)
We KNOW who commits the most crime... (Score:1)
... in London... and in which parts...
But we're not allowed to say...
The act of detecting changes your results (Score:1)
Or do you intend to have the cops lay low so they can "catch them in the act" or at least catch them quicker "after the fact"?
Re: (Score:2)
As for public data being collectively aggregated without permission - that's another story.
Hell, they should be handing out cell phones for free they use your data for so much nowadays.
Re: (Score:2)
ie not a gov ground station getting domestic calls.
UK law enforcement and political parties where more interested in phone calls, later cell phone tracking, rapid decryption of consumer grade computer encryption and getting legally safe convictions in closed courts.
Government Technical Assistance Centre (GCHQ Technical Assis
Bad Analogy (Score:3)
Re: (Score:2)
If you have a small enough town with a small enough cell size, it should be blindingly obvious which handset IMSI numbers where usually in the area when a crime was committed.
With enough data, you can simply map out the handset IMSI of the most probable perpetrators. There were 5 instances of a street robbery, at night, and the only common denominator is IMSI xyz that has been in the vicinity and moving around the time of all 5 robberies. It either is a totally unlucky individual or the most likely suspect.
Sigh. (Score:2)
"significantly"
I do not think that word means what you think it means.
How many false positives though (Score:2)
TL;DR
Re ...Crime location updated rarely" - WRONG! (Score:1)
Algorithms are not hindered by wishful thinking (Score:2)
We know that people that commit crimes are much more often from certain social and cultural backgrounds. There are untold numbers of "anecdotal evidence" around, but we don't want that to be true. So we tell ourselves white lies, blame victims, discount hundreds of incidents as "anecdotal evidence", pinpoint the few cases outside the norm and fabricate elaborate excuses about why such and such were practically forced to commit crime. We are constantly telling ourselves how we are to blame for not paying eno