1
00:00:00,560 --> 00:00:03,140
Hey there it's Andre game with your quick tip.

2
00:00:03,140 --> 00:00:03,780
That is.

3
00:00:03,890 --> 00:00:07,160
Let's talk about correlation analysis.

4
00:00:07,160 --> 00:00:08,000
What does that mean.

5
00:00:08,270 --> 00:00:14,100
Well correlation analysis simply means which attributes have correlations.

6
00:00:14,150 --> 00:00:18,740
So let's say one column is correlated to another column.

7
00:00:18,890 --> 00:00:27,830
Let's say we're trying to sell our home and there's two columns one column is the size of the land that

8
00:00:27,830 --> 00:00:34,730
the house is on and the other column is the size of the House and the floor space.

9
00:00:34,730 --> 00:00:43,070
Now let's say that when we're analyzing our data we notice that pretty much all houses that have a large

10
00:00:43,250 --> 00:00:51,440
land size also have a large floor space size and they're just correlated the prices go up the same every

11
00:00:51,440 --> 00:00:59,220
time the land increases and every time the floor space increases in this case these attributes have

12
00:00:59,220 --> 00:01:01,890
high correlation with each other.

13
00:01:01,890 --> 00:01:09,240
In this case we can actually remove this from our analysis or from building our model because it might

14
00:01:09,240 --> 00:01:10,920
not affect our model.

15
00:01:10,920 --> 00:01:16,070
This gets to our next point or forward backwards attribute selection.

16
00:01:16,170 --> 00:01:17,980
What does this mean.

17
00:01:17,980 --> 00:01:26,280
Well we can try training our model using different techniques backward attributes selection essentially

18
00:01:26,280 --> 00:01:33,840
says train the model on all the attributes and then slowly to start taking away attributes or columns

19
00:01:34,200 --> 00:01:35,880
to train your model.

20
00:01:36,090 --> 00:01:43,150
It does that affect your model does it improve the model forward attribute selection is the opposite.

21
00:01:43,200 --> 00:01:50,460
Start with just one column when you train the model and keep adding one attribute at a time until you

22
00:01:50,460 --> 00:01:53,010
get the accuracy to plateau.

23
00:01:53,100 --> 00:02:00,400
That is if you keep increasing columns and let's say after the fiftieth column all the other attributes

24
00:02:00,400 --> 00:02:02,630
say you added just don't improve the model.

25
00:02:02,700 --> 00:02:04,710
Well then maybe we might not need it.

26
00:02:04,950 --> 00:02:07,120
This idea of correlation analysis.

27
00:02:07,200 --> 00:02:13,950
The forward backward attribute selection are all ways for us to test our model reduce our data if we

28
00:02:13,950 --> 00:02:21,120
want to and play with our model instead of just assuming if we include everything everything will make

29
00:02:21,120 --> 00:02:22,320
the model better.

30
00:02:22,320 --> 00:02:25,640
That's often not usually the case.

31
00:02:25,650 --> 00:02:26,290
All right.

32
00:02:26,370 --> 00:02:27,390
Back to Daniel.