Lets say you live in an apartment complex. In the past month, a number of
vehicles parked in the unsecured apartment complex have been broken into.
Items stolen included laptops, credit cards, purses, wallets, stereo
equipment etc.
Of those reported....some folks don't bother.....you have a list of the
dates and times that the victims discovered and reported the crime. Times may
or may not be accurate, because the victim might have been asleep and didn't
discover they were a victim until the next morning. The apartment residents
want some action and help to catch the criminal and you need to advise the
boss when would be a good time to send undercover officers to the area to
observe activities in the parking lot.

The dates and times are not necessarily evenly spaced.
Days vary from 1 to 5, 10,20 between incidents.
Times vary within the 24 hour clock.
However, you can see patterns in the data, i.e. the time may be between 1AM
and 5AM, sometimes earlier or later but the majority are in one time frame.
Days could be say always Wednesday or maybe Sunday and Thursday, or Tuesday,
Wednesday and Thursday.
There is a pattern there.
So all I'm trying to do is find the formula(s) that will give the best
probability of when the next incident will occur based on the days between
past events and the time of day that past events occured.

I know it can be done.....but not how to do it.
Rich
"SteveW" wrote:

No, seriously you will need some sort of curve fit
First step would be to use charting and try various
curve fits.
Use Line chart, then add trendline
Probably polynomial
But this is the thing, you can fit *any* curve through 3, 4
or 5 points, especially if the points are evenly spaced.

The problem is that this would surely only mean that all future
points on the line are potential *projection dates*

Steve



On Wed, 07 Feb 2007 03:25:00 -0000, WR1CH
wrote:

In other words you don't have a "clue".

"SteveW" wrote:

I think you'd be better posting on a forum where the bad guys are
They obviously have the dates/times in a spreadsheet
and plan there future crimes accordingly :)

Steve

On Wed, 07 Feb 2007 03:09:00 -0000, WR1CH
wrote:

Steve
Thanks for the response, actually the times are important as well.
In trying to predict a pattern for a crime...like burglary, we not
only
use
date and time of day that it happened but location as well. Of course
the
location is mapped.
We need both the date and time to predict what day and what time of
day
to
patrol or stake out a potential area that a burglary, robbery or ? may
take
place.
So I am trying to predict the next earliest and latest date and the
time
range within those dates that the incident may likely occur. That way
police
resources can be deployed with some statistical probability of
success....hopefully.
Rich

"SteveW" wrote:

Your best bet is to fit a curve to the 3 or more values
I'd ignore the times, if the dates are not too close

The advantage of a chart, is that as more data is added
the accuracy of the chart can be seen.

Much better of course if you weren't using crimes as data :)

Steve

On Tue, 06 Feb 2007 19:57:00 -0000, WR1CH
wrote:

Mike
I am trying to predict the next date/time period that another crime
will
occur based on the periods between the dates/times that have
already
occured.
Thanks

"Mike" wrote:

P(a)=No ways event can happen/No possible outcomes.

Take throwing a 6 with a standard dice.

P(a)= 1/6

You need to be a lot more specific about what you are trying to
predict
if
you want to incorporate date/time into the equation.

"WR1CH" wrote:

Trying to find the formulas to use to project the next probable
occurance of
an event based on date and time....2 separate fields date in one
time
in the
other.
The events usually 3 or more.