1
00:00:00,100 --> 00:00:00,900
Hello my friends.

2
00:00:00,900 --> 00:00:01,600
Welcome back.

3
00:00:01,600 --> 00:00:06,900
And mostly welcome to part four, making
the predictions and evaluating the model.

4
00:00:07,233 --> 00:00:10,600
And at the end of the previous tutorial,
I actually asked you to do a little

5
00:00:10,633 --> 00:00:12,600
homework, which is right here,

6
00:00:12,600 --> 00:00:16,666
and which consisted
of using your Ein and model to predict

7
00:00:16,666 --> 00:00:21,300
if the customer with the following
information will leave yes or no.

8
00:00:21,300 --> 00:00:22,200
The bank.

9
00:00:22,200 --> 00:00:23,633
So I hope you did well.

10
00:00:23,633 --> 00:00:26,033
You actually had all the tools to do it.

11
00:00:26,033 --> 00:00:29,033
And that's something we did before in part
three classification.

12
00:00:29,166 --> 00:00:30,966
So I'm sure you smashed this.

13
00:00:30,966 --> 00:00:32,033
Let's see about that.

14
00:00:32,033 --> 00:00:35,033
Let's implement the solution together.

15
00:00:35,266 --> 00:00:35,600
All right.

16
00:00:35,600 --> 00:00:38,166
So I left this code
cell here to enter the solution.

17
00:00:38,166 --> 00:00:39,300
But this is just a common.

18
00:00:39,300 --> 00:00:41,333
So let's just remove it.

19
00:00:41,333 --> 00:00:45,000
And now how do we complete
successfully that homework.

20
00:00:45,000 --> 00:00:49,433
Well of course we need to get first
or a and model from which

21
00:00:49,466 --> 00:00:51,200
we're going to call a new method.

22
00:00:51,200 --> 00:00:54,900
You know, different than the add method
or the fit method.

23
00:00:55,300 --> 00:00:59,133
Right now of course,
we need to call the predict method right.

24
00:00:59,133 --> 00:01:00,633
Absolutely no trap.

25
00:01:00,633 --> 00:01:03,666
We call this break method to predict
if yes or no.

26
00:01:03,666 --> 00:01:06,866
The customer with the following
informations will leave the bank.

27
00:01:07,366 --> 00:01:07,800
All right.

28
00:01:07,800 --> 00:01:11,966
And now since this is a method
I'm going to add some parentheses here.

29
00:01:12,233 --> 00:01:13,500
Let me actually scroll down.

30
00:01:13,500 --> 00:01:17,433
You know in case you're uncomfortable
with the border at the bottom.

31
00:01:17,566 --> 00:01:19,733
All right. So predict then okay.

32
00:01:19,733 --> 00:01:22,200
So you have all the information
of the predict method.

33
00:01:22,200 --> 00:01:24,766
But we know exactly what to predict.

34
00:01:24,766 --> 00:01:27,900
This is given
in the instructions of the homework.

35
00:01:28,166 --> 00:01:30,233
We have to enter these informations.

36
00:01:30,233 --> 00:01:32,800
But there are a few things
to pay attention to.

37
00:01:32,800 --> 00:01:34,100
And there are actually three things.

38
00:01:34,100 --> 00:01:36,100
That's why
I really wanted to give you this homework,

39
00:01:36,100 --> 00:01:39,366
because I wanted to see
if you could pay attention

40
00:01:39,366 --> 00:01:42,933
to three important things
to include inside that predict method.

41
00:01:43,400 --> 00:01:45,966
Let's start with the most essential one,

42
00:01:45,966 --> 00:01:49,666
which is the double
pair of square brackets.

43
00:01:49,833 --> 00:01:50,300
Right.

44
00:01:50,300 --> 00:01:55,366
It's very important to remember that
any input of the predict method

45
00:01:55,500 --> 00:01:58,666
must be a 2D array,
whether it is the test set

46
00:01:59,133 --> 00:02:02,433
to predict an in symbol of outcomes
for an in

47
00:02:02,433 --> 00:02:06,600
symbol of observations,
or whether it is a single prediction.

48
00:02:06,600 --> 00:02:10,466
Well, anything has to be in a double pair
of square brackets,

49
00:02:10,466 --> 00:02:12,866
which therefore makes a 2D array.

50
00:02:12,866 --> 00:02:17,566
Basically, the format expected
by the break method is always a 2D array.

51
00:02:18,100 --> 00:02:19,700
Okay, so that was the first thing.

52
00:02:19,700 --> 00:02:21,633
Then inside this 2D array

53
00:02:21,633 --> 00:02:24,633
we're going to enter
well the different informations here.

54
00:02:25,000 --> 00:02:28,433
And that leads me to mention
the second important thing

55
00:02:28,433 --> 00:02:33,000
which you had to pay attention
to which is about that variable here.

56
00:02:33,000 --> 00:02:35,633
Geography France.

57
00:02:35,633 --> 00:02:38,233
According to you, do we have to enter,

58
00:02:38,233 --> 00:02:41,400
as you know, this first information here
for the geography variable,

59
00:02:41,600 --> 00:02:45,700
do we have to enter the string France
or do we have to enter something else?

60
00:02:45,900 --> 00:02:49,133
Well, of course we have to enter
something else and that's something else.

61
00:02:49,133 --> 00:02:52,233
Or the values of the dummy variables.

62
00:02:52,500 --> 00:02:55,200
Right. That's the second thing
you had to do correctly.

63
00:02:55,200 --> 00:02:59,700
And therefore now we need to check
what was the encoding for France.

64
00:02:59,700 --> 00:03:03,133
Well remember if we have a look
at our matrix of features

65
00:03:03,133 --> 00:03:06,400
X right above,
you know important data preprocessing.

66
00:03:06,400 --> 00:03:11,300
So this is the matrix of features X, the
whole matrix containing all the customers.

67
00:03:11,300 --> 00:03:13,833
So in order to know
what France corresponds to

68
00:03:13,833 --> 00:03:17,233
you know, in terms of encoding,
we just have to have a look

69
00:03:17,233 --> 00:03:21,866
at the first observation
before one hot encoding was applied.

70
00:03:22,166 --> 00:03:25,166
Because indeed that first observation
corresponds to France.

71
00:03:25,500 --> 00:03:27,533
And so now we just need to compare
friends here

72
00:03:27,533 --> 00:03:31,666
to the one hot encoding resulting
from what we did here.

73
00:03:31,666 --> 00:03:35,000
And well, we can see that
the first row here contains

74
00:03:35,000 --> 00:03:38,766
100 as the values of the dummy variables.

75
00:03:38,766 --> 00:03:42,700
And therefore France was encoded
indeed into 100.

76
00:03:42,700 --> 00:03:45,866
So the dummy variable values
for the France country

77
00:03:45,866 --> 00:03:49,433
and the geography variable is indeed 100.

78
00:03:49,433 --> 00:03:55,500
And that's exactly what we have to enter
here as the first parameter.

79
00:03:55,500 --> 00:03:56,300
So let's do this.

80
00:03:56,300 --> 00:04:00,500
Let's enter 100 and there we go.

81
00:04:00,500 --> 00:04:05,200
Basically that enter is that first value
of the first parameter geography.

82
00:04:06,033 --> 00:04:06,500
All right.

83
00:04:06,500 --> 00:04:08,966
And then all the rest is easy.

84
00:04:08,966 --> 00:04:11,966
For the credit score
we will enter indeed 600.

85
00:04:12,566 --> 00:04:15,133
Then let's see the gender male.

86
00:04:15,133 --> 00:04:19,133
So remember male was encoded
as one and female was encoded as zero.

87
00:04:19,133 --> 00:04:22,733
So here in order to enter the male gender
we need to enter one.

88
00:04:23,200 --> 00:04:25,766
All right then 40 years old very simply

89
00:04:25,766 --> 00:04:29,666
we enter 40 then tenure three years.

90
00:04:29,666 --> 00:04:31,933
We enter three here.

91
00:04:31,933 --> 00:04:34,533
Then balance $60,000.

92
00:04:34,533 --> 00:04:37,533
So we enter 60,000.

93
00:04:37,700 --> 00:04:40,466
Then let's see number of products two.

94
00:04:40,466 --> 00:04:42,133
So two.

95
00:04:42,133 --> 00:04:45,366
Then does
this customer have a credit card. Yes.

96
00:04:45,766 --> 00:04:49,666
So one because one corresponds to yes
and zero corresponds to no.

97
00:04:50,033 --> 00:04:52,566
Is this customer an active member. Yes.

98
00:04:52,566 --> 00:04:57,866
So once again
an estimated salary finally $50,000.

99
00:04:57,966 --> 00:05:03,700
So our last parameter
value here is $50,000 okay.

100
00:05:04,200 --> 00:05:05,833
Yes. All good.

101
00:05:05,833 --> 00:05:06,700
Excellent.

102
00:05:06,700 --> 00:05:08,966
And now I have a question for you.

103
00:05:08,966 --> 00:05:11,766
Do you think we're done with this
prediction.

104
00:05:11,766 --> 00:05:14,766
You know using the predict method
with all these informations.

105
00:05:15,300 --> 00:05:18,166
And the answer to that is no.

106
00:05:18,166 --> 00:05:22,233
And that leads me to the third thing
you had to pay attention to, which is

107
00:05:22,233 --> 00:05:25,433
the fact that remember to predict

108
00:05:25,433 --> 00:05:28,933
method should be called under observations

109
00:05:29,133 --> 00:05:34,233
onto which the same scaling was applied
as in the training.

110
00:05:34,500 --> 00:05:40,033
And since we trained our artificial neural
network with the scaled values,

111
00:05:40,033 --> 00:05:40,300
you know,

112
00:05:40,300 --> 00:05:45,000
the scaled values of the features as well,
the predict method has to be called

113
00:05:45,000 --> 00:05:49,966
on to these observations
to which the same scaling was applied.

114
00:05:50,233 --> 00:05:54,633
And that was the third thing you must have
not forgotten, which is the fact

115
00:05:54,633 --> 00:06:00,500
that you have to apply your object here
to scale that single observation, right?

116
00:06:00,500 --> 00:06:01,800
That's super important.

117
00:06:01,800 --> 00:06:06,200
Make sure to pay attention to this check
if some scaling was applied

118
00:06:06,200 --> 00:06:08,400
during the training. And yes,
it is the case here.

119
00:06:08,400 --> 00:06:11,400
And anyway,
it is always the case for neural networks.

120
00:06:11,500 --> 00:06:13,166
And therefore in the predict method.

121
00:06:13,166 --> 00:06:16,166
Well we need to scale
that single observation.

122
00:06:16,166 --> 00:06:19,800
And the way to do
this is of course by calling our subject

123
00:06:20,333 --> 00:06:24,000
yes, this one
from which we're going to call be careful

124
00:06:24,000 --> 00:06:27,066
not the fit transform,
because the fit transform is used

125
00:06:27,066 --> 00:06:31,333
to remember, get the mean
and the standard deviation of the values

126
00:06:31,333 --> 00:06:35,133
in the training set in order
to fit your scaler to the training set.

127
00:06:35,133 --> 00:06:39,233
But then for the test set,
we only need to call the transform method,

128
00:06:39,433 --> 00:06:41,766
because if we fit it again to scaler,

129
00:06:41,766 --> 00:06:44,400
well that would cause some information
leakage.

130
00:06:44,400 --> 00:06:47,133
You know, I explained this in part
one data preprocessing.

131
00:06:47,133 --> 00:06:50,100
Check it out again
if you need, but remember that

132
00:06:50,100 --> 00:06:53,366
on the test set or on new observations
on which you deploy

133
00:06:53,433 --> 00:06:57,600
your model in production,
you can only apply the transform method.

134
00:06:57,900 --> 00:07:00,900
And that's what we're going to call here
transform them.

135
00:07:01,333 --> 00:07:01,933
There we go.

136
00:07:01,933 --> 00:07:03,866
Which has to take as input

137
00:07:03,866 --> 00:07:07,800
the whole observation
in these double pair of square brackets.

138
00:07:07,966 --> 00:07:08,766
All right.

139
00:07:08,766 --> 00:07:10,700
Perfect good good good.

140
00:07:10,700 --> 00:07:12,833
So basically that's it.

141
00:07:12,833 --> 00:07:14,200
Now we can run that cell.

142
00:07:14,200 --> 00:07:16,133
Now we can get our prediction.

143
00:07:16,133 --> 00:07:19,266
But remember that when compiling our.

144
00:07:19,266 --> 00:07:23,033
And to you know enough to miser
a loss function and the metrics.

145
00:07:23,166 --> 00:07:26,966
Remember that in the output
of our artificial neural network,

146
00:07:26,966 --> 00:07:32,233
we chose a sigmoid activation function,
which therefore will return

147
00:07:32,233 --> 00:07:35,233
this prediction
in the form of a probability,

148
00:07:35,233 --> 00:07:38,233
meaning that
instead of getting the final outcome

149
00:07:38,366 --> 00:07:42,000
1 or 0, whether the customer left
or stayed in the bank,

150
00:07:42,200 --> 00:07:47,466
well, we're going to get the probability
that this customer leaves the bank.

151
00:07:47,700 --> 00:07:51,700
But you'll see it will be very easy
to get the final prediction 1 or 0.

152
00:07:51,700 --> 00:07:54,433
But first let me show you what we get
with this prediction.

153
00:07:54,433 --> 00:07:58,733
So we're going to put all this
into a print right.

154
00:07:58,800 --> 00:08:01,800
All this into a third pair of parenthesis.

155
00:08:02,100 --> 00:08:04,533
Now let me play that cell.

156
00:08:04,533 --> 00:08:09,466
And we get that,
that the predicted probability

157
00:08:09,466 --> 00:08:13,800
that discussed and released to bank
is 0.04,

158
00:08:13,800 --> 00:08:16,800
you know, 0.0 38, 0.04.

159
00:08:17,000 --> 00:08:19,000
And therefore it is predicted

160
00:08:19,000 --> 00:08:22,800
that this customer has a very low chance
to leave the bank.

161
00:08:23,266 --> 00:08:27,100
And if you don't want the outcome
in the form of a probability, well,

162
00:08:27,100 --> 00:08:31,066
the trick to convert this into
the final prediction is very easy.

163
00:08:31,500 --> 00:08:35,266
You just need to, as here, you know,
just before the last parenthesis,

164
00:08:35,733 --> 00:08:39,166
the larger symbol and then 0.5.

165
00:08:39,333 --> 00:08:40,466
Why is that?

166
00:08:40,466 --> 00:08:42,466
That's because you know all this.

167
00:08:42,466 --> 00:08:44,666
You know and predict the next transform.

168
00:08:44,666 --> 00:08:47,766
Then all the informations of the customers
up to here

169
00:08:47,966 --> 00:08:51,500
returns exactly the predicted probability.

170
00:08:51,900 --> 00:08:57,833
And here we choose a threshold of 0.5
to say that if that predicted probability

171
00:08:57,833 --> 00:09:02,766
is larger than 0.5, well, we will consider
the final result to be one.

172
00:09:02,766 --> 00:09:03,133
Right.

173
00:09:03,133 --> 00:09:07,766
Because the predicted probability that
the outcome is one is larger than 0.5,

174
00:09:07,766 --> 00:09:11,700
meaning that there is more than 50% chance
that the predicted outcome is one.

175
00:09:11,700 --> 00:09:13,200
So we'll consider it to be one.

176
00:09:13,200 --> 00:09:16,800
And however, if the predicted probability
that the customer leaves

177
00:09:16,800 --> 00:09:21,400
the bank is below 0.5, well,
we will consider it to be zero.

178
00:09:21,500 --> 00:09:24,033
Of course, you can choose
a different value of the threshold,

179
00:09:24,033 --> 00:09:27,733
especially when you have critical outcome,
you know, for a critical decision.

180
00:09:27,933 --> 00:09:31,166
And in that case you can, for example,
increase the threshold.

181
00:09:31,166 --> 00:09:34,133
But here let's just pick
the default value 0.5.

182
00:09:34,133 --> 00:09:36,866
This totally makes sense
for our case study.

183
00:09:36,866 --> 00:09:40,533
And therefore
now if we re-execute this cell, we will

184
00:09:40,533 --> 00:09:44,433
of course get the final outcome
which will be of course zero or false.

185
00:09:44,433 --> 00:09:45,633
You know, false.

186
00:09:45,633 --> 00:09:48,600
We should not say goodbye to that customer
or false.

187
00:09:48,600 --> 00:09:51,066
You know, 
the customer won't leave the bank.

188
00:09:51,066 --> 00:09:53,066
In other words, it will stay in the bank.

189
00:09:53,066 --> 00:09:55,133
So good, good news for the bank.

190
00:09:55,133 --> 00:09:59,633
But anyway, I just wanted to show you
how to use your artificial neural network

191
00:09:59,633 --> 00:10:03,700
to predict the outcome of a single
observation, meaning a single customer.

192
00:10:03,900 --> 00:10:05,233
Remember the three things.

193
00:10:05,233 --> 00:10:07,133
First, you have to enter your informations

194
00:10:07,133 --> 00:10:10,366
in a 2D array
with a double pair of square brackets.

195
00:10:10,600 --> 00:10:13,833
Then you need to be careful
with the dummy variable values here,

196
00:10:13,833 --> 00:10:17,900
instead of entering the string input
for that categorical variable.

197
00:10:18,033 --> 00:10:22,300
And then the third thing remember
to apply the scaling from your subject

198
00:10:22,300 --> 00:10:27,466
because your artificial neural network
was trained with scaled values okay.

199
00:10:27,466 --> 00:10:28,966
Great. All right.

200
00:10:28,966 --> 00:10:32,300
So now let's finish this implementation
in a flashlight.

201
00:10:32,300 --> 00:10:34,400
You know the last two cells we have.

202
00:10:34,400 --> 00:10:37,466
By the way I gave you some important notes
here so that you can remember.

203
00:10:37,466 --> 00:10:39,100
That's basically what I explained.

204
00:10:39,100 --> 00:10:42,200
But anyway now we have two cells to go.

205
00:10:42,200 --> 00:10:45,866
The first one is, you know, that cell
which you know very well where we predict

206
00:10:46,000 --> 00:10:48,933
the test results
by displaying next to each other

207
00:10:48,933 --> 00:10:52,800
the vector of predictions widespread
and a vector of real results y test.

208
00:10:52,966 --> 00:10:56,000
So we're going to do this
the most efficiently by juggling

209
00:10:56,000 --> 00:10:57,033
with our toolkit.

210
00:10:57,033 --> 00:10:59,000
And finally same will make the confusion

211
00:10:59,000 --> 00:11:02,700
matrix
to get the final accuracy on the test set.

212
00:11:03,300 --> 00:11:04,833
All right let's do this efficiently.

213
00:11:04,833 --> 00:11:07,666
So we're going to go back into our machine
learning.

214
00:11:07,666 --> 00:11:10,200
Is it for there
you know to juggle with our toolkit.

215
00:11:10,200 --> 00:11:11,866
So I'm clicking this shortcut here

216
00:11:11,866 --> 00:11:15,566
to get the original base
of the folder machine learning A to Z.

217
00:11:16,100 --> 00:11:19,966
Then I'm just going to go, you know
in 2.3 classification to take any model.

218
00:11:19,966 --> 00:11:23,866
Let's take, you know, logistic regression
for example then Python.

219
00:11:24,133 --> 00:11:29,133
Then I can open this implementation
which contains indeed a lot of tools, you

220
00:11:29,133 --> 00:11:32,133
know, which we can do with classification

221
00:11:32,233 --> 00:11:35,166
as we are doing right now with R and N.

222
00:11:35,166 --> 00:11:38,533
And now I'm just going to scroll down
and get, you know, the final cells,

223
00:11:38,866 --> 00:11:42,133
which will do exactly
what we're about to do now with our an N

224
00:11:42,533 --> 00:11:45,900
first to display next to each other

225
00:11:46,333 --> 00:11:48,966
the vector of predictions
and the vector of real results.

226
00:11:48,966 --> 00:11:50,600
So let's get that cell.

227
00:11:50,600 --> 00:11:55,100
Let's base
that inside a new code cell right here.

228
00:11:55,500 --> 00:12:00,500
But remember that now since our classifier
which is no longer called classifier

229
00:12:00,500 --> 00:12:05,566
but a and n, returns the predictions
in the form of probabilities,

230
00:12:05,800 --> 00:12:08,733
well, we just have to do a little
something here, which is to

231
00:12:08,733 --> 00:12:12,333
of course
convert our predicted probabilities

232
00:12:12,333 --> 00:12:16,700
which are in white bread,
into predicted binary outcomes 001.

233
00:12:16,800 --> 00:12:21,466
And the trick to do that is exactly
the same as the trick we used here.

234
00:12:21,466 --> 00:12:25,800
It is, of course, to update y pred
the following way

235
00:12:26,133 --> 00:12:32,000
by setting it equal to well,
the boolean result of whether y pred.

236
00:12:32,000 --> 00:12:36,933
You know, the predicted
probability is larger than 0.5,

237
00:12:36,933 --> 00:12:41,166
so that if y pred is between zero and 0.5,

238
00:12:41,400 --> 00:12:44,900
then this new y bread will become zero,
because this won't be true.

239
00:12:45,166 --> 00:12:47,366
And if y print is larger than 0.5,

240
00:12:47,366 --> 00:12:50,866
then this will be true, and therefore
y bread will be equal to one.

241
00:12:51,100 --> 00:12:51,500
All right.

242
00:12:51,500 --> 00:12:55,500
And we'll get therefore
the final predicted binary outcomes 001.

243
00:12:56,033 --> 00:12:57,400
And then the rest is the same.

244
00:12:57,400 --> 00:12:58,700
All this is good.

245
00:12:58,700 --> 00:13:04,500
So let's play that cell and we'll
get indeed displayed next to each other.

246
00:13:04,733 --> 00:13:07,733
First on the left
a vector of predictions y pred,

247
00:13:07,866 --> 00:13:11,366
and on the right
a vector of real results y test.

248
00:13:11,700 --> 00:13:14,166
And now we don't have all the results
displayed.

249
00:13:14,166 --> 00:13:16,266
That's
because we have a lot of observations.

250
00:13:16,266 --> 00:13:19,400
But you know, we can already see that
the predictions look pretty good.

251
00:13:19,700 --> 00:13:22,366
That first customer of the test set.
Be careful.

252
00:13:22,366 --> 00:13:24,466
That's all the customers of the test set.

253
00:13:24,466 --> 00:13:26,733
Well, that first customer in reality

254
00:13:26,733 --> 00:13:30,300
stayed in the bank and was predicted
indeed to stay in the bank.

255
00:13:30,600 --> 00:13:33,266
However, that second customer
actually left.

256
00:13:33,266 --> 00:13:38,300
In reality, the bank but was predicted
not to was predicted to stay in the bank.

257
00:13:38,700 --> 00:13:41,700
Then the third customer stayed in reality

258
00:13:41,700 --> 00:13:44,700
in the bank and was predicted
to stay in the bank as well.

259
00:13:44,833 --> 00:13:49,200
And then for these three last customers
of the test set, well, they in reality

260
00:13:49,200 --> 00:13:52,600
stayed in the bank and were predicted
to stay indeed in the bank.

261
00:13:52,600 --> 00:13:54,600
So the results look really good.

262
00:13:54,600 --> 00:13:59,200
But the real way to check
this is with our confusion matrix.

263
00:13:59,200 --> 00:14:03,700
And therefore here in our logistic
regression tool, you know, code template.

264
00:14:03,833 --> 00:14:07,933
Well, we just have to scroll down
a little bit more to find the code

265
00:14:07,933 --> 00:14:12,300
for the confusion matrix
and also the computation of the accuracy.

266
00:14:12,600 --> 00:14:17,366
So let's just get this then
let's just copy that into a new code cell.

267
00:14:17,833 --> 00:14:18,900
Let's just paste it.

268
00:14:18,900 --> 00:14:22,200
And now
are you ready to get the final accuracy

269
00:14:22,200 --> 00:14:26,133
of your very first artificial
neural network on the test set right.

270
00:14:26,133 --> 00:14:29,600
Meaning a new customer
on which the model wasn't trained?

271
00:14:29,833 --> 00:14:31,500
Well, let's check this out right now.

272
00:14:31,500 --> 00:14:34,833
Let's play that sale. And very good.

273
00:14:35,000 --> 00:14:38,966
We get an accuracy over 86%.

274
00:14:38,966 --> 00:14:42,866
That's really good because it means that
out of 100 customers,

275
00:14:43,000 --> 00:14:48,333
well, 86 were predicted correctly
to either stay in or leave the bank.

276
00:14:48,566 --> 00:14:51,600
And we can actually see here
the details of these predictions.

277
00:14:51,933 --> 00:14:54,733
We had 1520

278
00:14:54,733 --> 00:14:57,733
correct predictions
that the customer stays in the bank.

279
00:14:57,900 --> 00:15:02,333
Then we had 203 correct predictions
that the customer leaves the bank.

280
00:15:02,633 --> 00:15:07,166
Then we have 75 incorrect predictions
that the customer leaves the bank,

281
00:15:07,400 --> 00:15:12,000
and 202 incorrect predictions
that the customer stays in the bank.

282
00:15:12,533 --> 00:15:13,966
So anyway, that looks pretty good.

283
00:15:13,966 --> 00:15:17,533
The accuracy of 86%
is a very decent accuracy.

284
00:15:17,833 --> 00:15:20,133
Of course, I challenge you to beat that.

285
00:15:20,133 --> 00:15:22,000
You know, to at least get 1 or 2

286
00:15:22,000 --> 00:15:26,100
extra percent of accuracy,
but I actually experimented in myself

287
00:15:26,100 --> 00:15:30,066
and it will be pretty hard to beat this,
so don't waste too much time on this.

288
00:15:30,300 --> 00:15:34,733
Plus, I rather encourage you to, you know,
check out this free course that we have

289
00:15:34,733 --> 00:15:37,200
on Artificial Neural Network
for regression,

290
00:15:37,200 --> 00:15:40,466
because I'd really like you to do anything
you want with

291
00:15:40,466 --> 00:15:43,733
artificial neural networks,
whether it is to do classification

292
00:15:43,733 --> 00:15:49,033
for either a binary outcome
or a non binary outcome or regression.

293
00:15:49,200 --> 00:15:49,666
Right.

294
00:15:49,666 --> 00:15:51,633
With this course
plus the other free course,

295
00:15:51,633 --> 00:15:55,366
you'll know how to do anything
with artificial neural networks.

296
00:15:55,366 --> 00:15:57,733
So that's rather my recommendation.

297
00:15:57,733 --> 00:16:01,100
And now of course
I want to say a final congrats Lations.

298
00:16:01,266 --> 00:16:04,966
You completed a very advanced
branch of machine learning, so

299
00:16:04,966 --> 00:16:07,333
you can really be proud of your progress.

300
00:16:07,333 --> 00:16:11,200
And if you're not too overwhelmed, well,
it's time to reach an even higher level

301
00:16:11,200 --> 00:16:14,200
with convolutional Neural networks.

302
00:16:14,200 --> 00:16:17,766
So as soon as you feel ready for it, well,
join us, Kiril

303
00:16:17,766 --> 00:16:22,500
and I, in the next section
to tackle convolutional neural networks,

304
00:16:22,766 --> 00:16:24,966
and until then, enjoy machine learning.