WEBVTT 00:00.300 --> 00:10.450 OK, so now let's have a look at the other solution, the solution for the gaming's flustering project. 00:10.740 --> 00:15.530 So this project is basically named Bastin Market Segmentation. 00:15.840 --> 00:20.750 So we will be using K means clustering for this particular project. 00:20.760 --> 00:23.280 You can use any other clustering algorithm. 00:25.050 --> 00:34.700 So here in drafting the user, basically the dataset is representing a random sample of thirty thousand 00:34.710 --> 00:43.500 US high school students who had profiles on the well-known S.A. So to protect the users anonymity, 00:43.500 --> 00:51.510 the census will remain unnamed and the data was sampled evenly across four different high school graduation 00:51.510 --> 00:59.460 years, representing the senior, junior, sophomore and freshman classes at the time of the data collection. 01:00.450 --> 01:08.880 Now, this particular dataset contains 40 variables like gender, age range, basketball, football, 01:08.880 --> 01:09.360 soccer. 01:09.360 --> 01:17.460 All these different criteria in which the students might have interest in the final data setting basically 01:17.460 --> 01:24.330 indicate for each person how many times each word has appeared in their Ascendis profile. 01:24.840 --> 01:30.510 So now what we will be doing is the first thing which we do is we import all the important libraries 01:30.510 --> 01:31.710 and we load the details. 01:31.740 --> 01:42.390 And so here you can see the dataset contains graduation year, the gender, age number of friends and 01:42.390 --> 01:48.430 the words in which the person has interest rates and how many times it appears. 01:48.630 --> 01:51.780 So you can see here the friends of your sixty nine times. 01:52.950 --> 01:57.100 If you go further, you can see here this comes four times. 01:57.960 --> 02:01.410 So these are different kind of words which are coming in. 02:01.800 --> 02:12.220 Now let's give somebody statistics so you can see that the graduation year is ranging from 2006 to 2009. 02:13.080 --> 02:22.230 Age of the student is three years to one hundred and six years for a number of friends, ranges from 02:22.230 --> 02:24.560 zero to eight, 30 and so on. 02:24.990 --> 02:28.490 And you can see the standard deviation of different words. 02:29.050 --> 02:34.650 So here you can see that the meaning of the word, which you can see here, is both for basketball. 02:34.650 --> 02:36.000 It is 26. 02:37.350 --> 02:40.650 Then you can see these are a little less popular. 02:40.650 --> 02:42.600 Basketball seems to be highly popular. 02:42.630 --> 02:43.740 So is football. 02:44.340 --> 02:52.560 Then if you see, then this is a little more less of one. 02:52.580 --> 03:00.030 And then there are kisses and these are less popular and so on. 03:01.560 --> 03:04.510 So you can see how much these words are popular. 03:04.830 --> 03:06.760 So this basketball, this one. 03:06.790 --> 03:08.760 So these are point to six. 03:09.190 --> 03:17.970 Then you can see something which is more popular, like dance is more popular, zero point four two, 03:19.230 --> 03:26.820 then zero point seven three, which is very, very popular music zero point four six gaudens, very 03:26.820 --> 03:27.440 popular. 03:27.690 --> 03:35.550 So you can see these are different words which are so popular amongst these students and that's about 03:35.550 --> 03:35.670 it. 03:35.970 --> 03:41.990 So next, what we will be doing is you can see the gender, this differentiation. 03:42.240 --> 03:51.480 So there are total twenty seven thousand rules of data and out of which there are two different unique 03:51.480 --> 03:54.820 values and females, the majority here. 03:55.900 --> 03:57.260 Let's see the difference in. 03:57.770 --> 04:03.660 So you here you can see there are not much missing values, but there are around 5000 missing values 04:03.660 --> 04:04.410 for each. 04:04.440 --> 04:09.240 So what you can do is you can simply take the mean of the ages and those values. 04:09.900 --> 04:14.750 So here at or below five thousand records of missing ages. 04:14.790 --> 04:21.150 Also concerning is the fact that the minimum and maximum values seem to be invincible, that there's 04:21.600 --> 04:22.940 a minimum of three years. 04:22.960 --> 04:28.620 So that's not possible that a three year old would be using Essence's or someone our age one hundred 04:28.620 --> 04:30.420 and six would be attending high school. 04:30.780 --> 04:38.460 So what we simply do is we take the number of male and female candidate you can see around grindy two 04:38.460 --> 04:42.060 thousand female and there are only 5000 Miller there. 04:43.260 --> 04:48.310 And here's the two thousand gender values are missing. 04:48.690 --> 04:53.780 So you're not number twenty seven hundred. 04:54.120 --> 04:58.680 So here you can see that there are twenty two thousand female 5000 men. 04:59.010 --> 04:59.520 And when. 04:59.710 --> 05:06.640 Seven hundred missing values, so what we are trying going to do is we will fill all the null values 05:06.640 --> 05:07.730 with no gender. 05:08.260 --> 05:13.930 So we are saying that they have not disclosed their gender and we are not actually putting that as the 05:13.930 --> 05:20.490 majority value because then it would simply be female, because females there are already a majority. 05:20.500 --> 05:21.750 So we don't want to do that. 05:23.050 --> 05:28.620 So they put gender has not disclosed whatever the value is not a number. 05:30.040 --> 05:36.580 Next, what we do is we group the data based on the graduation year and the ages. 05:36.850 --> 05:38.490 So and we take the mean. 05:39.190 --> 05:40.990 So you can see the graduation. 05:40.990 --> 05:42.400 It is 2006. 05:42.820 --> 05:47.490 The age nine is in 19, then 18, 17, 16. 05:47.680 --> 05:53.680 So based on this graduation year, this seems that these are the reasonable ages which could be imputed. 05:53.950 --> 05:57.970 And this is a good way to actually find out what age a person should be. 05:58.360 --> 06:06.760 So what we will do is we will group the data by the graduation year and we will fill in the data based 06:06.760 --> 06:09.350 on the mean value from these values. 06:10.750 --> 06:19.330 So what we have got is we have imputed the values and similarly, we have found that moral values are 06:19.340 --> 06:20.140 present now. 06:20.140 --> 06:22.480 So we don't have any values present. 06:23.230 --> 06:26.130 Next, we will look at the outliers. 06:26.440 --> 06:32.770 So here you can see the original age range contains values from three to one hundred and six, which 06:32.770 --> 06:41.140 is unrealistic in nature because of age three or one hundred six would not entice a reasonable age range 06:41.140 --> 06:45.280 for people attending high school would be ranging from 13 to 21. 06:45.580 --> 06:52.180 The rest should be treated as outliers, keeping the age of student going to high school in mind so 06:52.180 --> 06:54.330 we will detect the outlier values. 06:54.330 --> 06:56.660 So we are generating the books plot. 06:56.710 --> 07:02.590 So here you can see that these are the actual correct values, the ones in the middle and all other 07:02.590 --> 07:04.440 values are actually outliers. 07:04.750 --> 07:12.670 So what we will do is we will find no the Q1 and Q2 values and accordingly we will find out the different 07:13.210 --> 07:15.790 ages for the outlier detection. 07:17.080 --> 07:25.320 And here you can see that twenty five percentile is 16 and maximum is 21. 07:25.330 --> 07:28.810 So what we are doing is we are putting the values here. 07:28.870 --> 07:31.200 What we have applied this condition. 07:31.660 --> 07:43.720 So what we have done, we have kept the data where the data age is greater than Q1 minus one point five 07:44.290 --> 07:48.640 and the data age is less than Q3 one point five IQ. 07:48.790 --> 07:51.190 So we have kept only these particular ages. 07:51.700 --> 07:56.820 So from this data now we will check the latest data, which we now have. 07:57.070 --> 08:07.420 So now we have reduced the number of rules to twenty nine thousand noley and now we have the minimum 08:07.420 --> 08:11.740 age as thirteen point seven one and the maximum age as 21 years. 08:11.750 --> 08:12.040 Now 08:15.190 --> 08:23.650 they now what we do is you can see now there are no outliers now regarding data. 08:23.650 --> 08:32.260 Preprocessing a common practice employed prior to any analysis using distance calculation is to normalize 08:32.260 --> 08:34.870 visa standardize for future. 08:35.110 --> 08:45.940 So we cannot have values with different ranges when we are using an algorithm which uses distance as 08:45.940 --> 08:47.200 its base metric. 08:48.160 --> 08:56.280 So all the distance based algorithms will need to have standard scanning applied. 08:56.350 --> 08:59.320 So here we will be applying scaling. 08:59.680 --> 09:02.220 So we will be applying standard schema here. 09:03.790 --> 09:11.560 So the process for ZAKES standardization skews the features so that they have a mean of zero one standard 09:11.560 --> 09:12.500 deviation of one. 09:12.880 --> 09:19.190 So this transformation changes the interpretation of data in a way that may be more useful. 09:19.510 --> 09:21.820 So that is what we will be doing. 09:23.740 --> 09:30.580 So here we are taking out the column names which have values which would be having different ranges. 09:32.630 --> 09:41.230 And we are applying standards killer on them, and now we have these skin granules present with us, 09:43.520 --> 09:49.580 so then next thing which will be done was to convert the objectivity to numerical. 09:49.610 --> 09:51.950 That is the categorical variables to numerical. 09:52.550 --> 09:55.080 So we will be doing the same. 09:55.100 --> 10:00.800 So we will take care of the gender converted into one two in three different categories. 10:01.970 --> 10:04.290 So now we have this particular data. 10:04.310 --> 10:10.590 After the transformation, the next thing which basically could be done is applying the key means model. 10:11.930 --> 10:16.090 So we have applied the gaming's model and we have for model. 10:16.430 --> 10:21.550 Next thing which we will be doing is we are running for a different number of clusters. 10:21.710 --> 10:31.100 So we are running it from cluster range one to 20 and we are applying it and we are running the Aluko. 10:31.100 --> 10:36.820 We are printing the L will go to see which number of clusters is actually helpful for us. 10:37.250 --> 10:43.100 Now, the location of a mean the plot is generally considered as an indicator of appropriate number 10:43.100 --> 10:43.800 of clusters. 10:44.270 --> 10:50.450 Here we have this about the size of five, which is here. 10:52.050 --> 11:00.210 So we will keep the number of clusters to we find that this key to be fight, so we will fit gaming's 11:00.210 --> 11:04.710 with case equal to five and these are the details which we obtain. 11:04.710 --> 11:07.210 And you can interpret the cluster sizes. 11:07.230 --> 11:14.490 You can find out the Szilard score and evaluate the model accordingly. 11:14.880 --> 11:24.540 You can also do this using any other than gordito of your choice and see how good values you can obtain 11:24.540 --> 11:25.140 out of this. 11:25.290 --> 11:26.610 That is completely up to you. 11:26.640 --> 11:28.690 This is just one of the implementations. 11:29.250 --> 11:29.850 Thank you.