1
00:00:04,500 --> 00:00:08,600
In this video, we'll discuss
how we can create visualizations

2
00:00:08,600 --> 00:00:12,490
that are used in
predictive policing models.

3
00:00:12,490 --> 00:00:15,430
In almost every
application, before we even

4
00:00:15,430 --> 00:00:17,980
consider a predictive
model, we should

5
00:00:17,980 --> 00:00:21,560
try to understand
the historical data.

6
00:00:21,560 --> 00:00:25,340
Many cities in the United
States and around the world,

7
00:00:25,340 --> 00:00:30,260
provide logs of reported crimes,
usually the time, location,

8
00:00:30,260 --> 00:00:32,640
and nature of the event.

9
00:00:32,640 --> 00:00:36,350
In this lecture, we'll use
data from the city of Chicago,

10
00:00:36,350 --> 00:00:38,810
in the United States,
about motor vehicle thefts.

11
00:00:41,400 --> 00:00:43,820
Given this data
on crimes, suppose

12
00:00:43,820 --> 00:00:45,950
we wanted to communicate
crime patterns

13
00:00:45,950 --> 00:00:48,560
over the course of
an average week.

14
00:00:48,560 --> 00:00:52,690
We could display daily crime
averages using a line graph,

15
00:00:52,690 --> 00:00:57,030
like the one shown here, but
this doesn't seem too useful.

16
00:00:57,030 --> 00:01:00,960
We can see that crime tends to
be higher on Saturday, but when

17
00:01:00,960 --> 00:01:04,400
on Saturday, and where?

18
00:01:04,400 --> 00:01:08,830
We could replace our x-axis
with the hour of the day

19
00:01:08,830 --> 00:01:11,780
and have a different line
for every day of the week

20
00:01:11,780 --> 00:01:15,760
to understand when crime
occurs in more detail.

21
00:01:15,760 --> 00:01:19,150
But this would be a jumbled mess
with seven lines, and probably

22
00:01:19,150 --> 00:01:21,150
very hard to read.

23
00:01:21,150 --> 00:01:24,660
We could instead use no
visualization at all,

24
00:01:24,660 --> 00:01:27,810
and instead present
information in a table,

25
00:01:27,810 --> 00:01:30,039
like the one shown here.

26
00:01:30,039 --> 00:01:33,650
For each hour and day,
we have the total number

27
00:01:33,650 --> 00:01:36,240
of crimes that occurred.

28
00:01:36,240 --> 00:01:39,030
This is a valid
representation of the data,

29
00:01:39,030 --> 00:01:43,460
but large tables of numbers can
be hard to read and understand.

30
00:01:43,460 --> 00:01:46,590
So how can we make the table
more interesting and usable?

31
00:01:49,210 --> 00:01:51,700
A great way to
visualize information

32
00:01:51,700 --> 00:01:55,550
in a two-dimensional
table is with a heat map.

33
00:01:55,550 --> 00:02:00,050
Heat maps visualize data
using three attributes.

34
00:02:00,050 --> 00:02:04,030
Two of the attributes are on
the x and y-axes, typically

35
00:02:04,030 --> 00:02:07,990
displayed horizontally
and vertically.

36
00:02:07,990 --> 00:02:12,400
The third attribute is
represented by shades of color.

37
00:02:12,400 --> 00:02:16,930
In this example, lower
values in the third attribute

38
00:02:16,930 --> 00:02:19,780
correspond to colors
closer to blue,

39
00:02:19,780 --> 00:02:22,280
and higher values in
the third attribute

40
00:02:22,280 --> 00:02:25,780
correspond to colors
closer to red.

41
00:02:25,780 --> 00:02:29,920
For example, the x-axis
could be hours of the day,

42
00:02:29,920 --> 00:02:32,910
the y-axis could be
days of the week,

43
00:02:32,910 --> 00:02:36,020
and the colors could correspond
to the amount of crime.

44
00:02:38,630 --> 00:02:42,000
In a heat map, we can pick
different color schemes

45
00:02:42,000 --> 00:02:46,150
based on the type of data to
convey different messages.

46
00:02:46,150 --> 00:02:49,560
In crime, a yellow
to red color scheme

47
00:02:49,560 --> 00:02:52,410
might be appropriate
because it can highlight

48
00:02:52,410 --> 00:02:55,230
some of the more
dangerous areas in red.

49
00:02:55,230 --> 00:02:59,760
Your eye is naturally drawn
to the red areas of the plot.

50
00:02:59,760 --> 00:03:03,380
In other applications,
both high and low values

51
00:03:03,380 --> 00:03:07,050
are meaningful, so having
a more varied color scheme

52
00:03:07,050 --> 00:03:09,180
might be useful.

53
00:03:09,180 --> 00:03:11,340
And in other applications,
you might only

54
00:03:11,340 --> 00:03:13,970
want to see cells
with high values,

55
00:03:13,970 --> 00:03:17,550
so you could use a gray scale to
make the cells with low values

56
00:03:17,550 --> 00:03:19,380
white.

57
00:03:19,380 --> 00:03:24,360
The x and y-axes in a heat map
don't need to be continuous.

58
00:03:24,360 --> 00:03:27,680
In our example, we have a
categorical or factor variable

59
00:03:27,680 --> 00:03:29,960
-- the day of the week.

60
00:03:29,960 --> 00:03:32,840
And we can even
combine a heat map

61
00:03:32,840 --> 00:03:35,079
with a geographical
map, which we'll

62
00:03:35,079 --> 00:03:37,660
discuss later in this lecture.

63
00:03:37,660 --> 00:03:40,030
This type of heat map
is frequently used

64
00:03:40,030 --> 00:03:46,100
in predictive policing to show
crime hot spots in a city.

65
00:03:46,100 --> 00:03:49,400
In this lecture, we'll use
Chicago motor vehicle theft

66
00:03:49,400 --> 00:03:53,120
data to explore patterns
of crime, both over days

67
00:03:53,120 --> 00:03:57,130
of the week, and over
hours of the day.

68
00:03:57,130 --> 00:04:00,880
We're interested in analyzing
the total number of car thefts

69
00:04:00,880 --> 00:04:03,470
that occur in any
particular hour

70
00:04:03,470 --> 00:04:07,420
of a day of the week
over our whole data set.

71
00:04:07,420 --> 00:04:10,280
In the next video,
we'll switch to R,

72
00:04:10,280 --> 00:04:12,390
and start by
creating a basic line

73
00:04:12,390 --> 00:04:16,890
plot counting the total
crime by the day of the week.