WEBVTT 00:00.570 --> 00:07.350 Hi, I hope you have worked on this first project, which is the house price prediction problem, the 00:07.350 --> 00:13.740 next project which we would work under supervised learning, is, again, a regression project which 00:13.740 --> 00:18.090 is based on the property hazard and the school prediction. 00:18.450 --> 00:23.610 So in this particular project, we will be using supervised learning and regression. 00:24.240 --> 00:26.970 And this is an industry level project. 00:26.980 --> 00:34.110 So the data center would be a little more typical in nature and a little complex in nature in comparison 00:34.110 --> 00:35.050 to the previous one. 00:35.730 --> 00:44.910 So here we would be given a mass dataset which contains around forty thousand rows of properties and 00:44.910 --> 00:46.650 thirty for various dots. 00:46.830 --> 00:49.680 That is thirty for various columns of data. 00:50.190 --> 00:57.890 So this particular dataset deals with a problem which is property hazard problem. 00:58.920 --> 01:06.750 So it says that there are different properties in a particular location and one particular organization 01:06.750 --> 01:12.260 is trying to do a domain Damiens of these properties. 01:12.270 --> 01:16.860 They are trying to restore the properties. 01:17.100 --> 01:24.720 Now, when this organization takes up the task of restoring this property, so there are certain hazards 01:24.720 --> 01:30.870 which are associated with the property which is being restored, there could be a case where the property 01:30.870 --> 01:34.550 has zero risk and it is completely fine to restore it. 01:34.800 --> 01:40.550 Well, there could be a few properties where it would be very difficult to restore the property and 01:40.560 --> 01:43.860 there could be different hazards which could be possible. 01:43.860 --> 01:48.690 Or maybe the company might lose a lot of money by investing in that property. 01:48.690 --> 01:52.160 And still, it would not be good enough for selling to someone else. 01:52.770 --> 01:58.180 So that is a different risk level which will be associated with these properties. 01:58.410 --> 02:04.890 So here we have around forty thousand eight hundred properties and could be four different task authority 02:04.890 --> 02:13.050 for different risk factors associated with different things to which they are creating a generalized 02:13.050 --> 02:14.800 single risk score. 02:15.810 --> 02:24.360 So this risk score is actually depending on these thirty four different dust right now. 02:25.080 --> 02:30.440 These are for taking up in different heavy equipment maintenance by the manual. 02:30.630 --> 02:34.110 Also, the manual crew is actually doing these performing these days. 02:34.500 --> 02:40.680 So each of these days have been given a hazard score, which is eventually used to decide the level 02:40.680 --> 02:46.350 of safety checks and caution while planning for the maintenance process of these properties. 02:47.280 --> 02:54.660 So your task here is to build a predictive model for predicting the hazards, given other information 02:54.660 --> 02:59.670 related to the main building, that school may have given these 30 foot columns which have different 02:59.670 --> 03:02.520 maintenance task and did different properties. 03:02.700 --> 03:08.820 And out of that, you need to make sure that you create a model which is able to find out accurate risk. 03:09.870 --> 03:15.540 Now, one thing to note here is that the risk score is not a continuous value. 03:15.990 --> 03:19.660 So what we have been dealing with till now are continuous values. 03:19.830 --> 03:21.880 So now where does this go? 03:22.560 --> 03:26.940 Should this be taken up as a classification problem or regression problem? 03:26.940 --> 03:34.640 But because these are actual numbers and a lot of numbers, so it's not fit to be a classification problem. 03:35.520 --> 03:39.390 That is why you will be using a regressive it. 03:39.630 --> 03:45.370 So, for example, what you can do is you can use the objective function, which is counterpoise. 03:45.890 --> 03:55.510 So counterpoised is actually will be helpful in finding out in applying regression back for the numbers. 03:55.530 --> 03:58.500 That is something which will not have the floating point values. 03:58.710 --> 04:03.240 So you will be dealing with something which is in a whole number or integer form. 04:04.250 --> 04:04.520 Right. 04:04.590 --> 04:06.000 So we can use that. 04:06.360 --> 04:11.580 Apart from that, this project for this project, you will be working with me. 04:11.690 --> 04:12.270 Absolutely. 04:12.960 --> 04:18.570 So what you can do is you can find out the mean absolute error for the modules which you have created 04:18.930 --> 04:26.730 and you can find out the score as one minus the mean absolute error divided by a five point four. 04:26.940 --> 04:30.800 You don't need to really think much about how this coding is happening. 04:31.740 --> 04:39.330 You just need to make sure that the score, which you get after this particular calculation comes out 04:39.330 --> 04:41.610 to be more than zero point five. 04:41.640 --> 04:45.840 And that would be a good model, which you would have generated. 04:46.920 --> 04:56.490 So, again, there is no specific rule or no specific threshold value that your model has to be top 04:56.490 --> 04:59.880 notch or should have an accuracy of this or. 05:00.020 --> 05:02.960 I told this, I mean, absolutely nothing like that. 05:03.860 --> 05:07.610 This is the minimal point that the school has to be at least 2.5. 05:08.360 --> 05:10.130 That is greater than zero point five. 05:10.520 --> 05:12.740 And apart from that, there is no limit. 05:12.740 --> 05:18.020 You can keep improving your model, implement different models, compare different models, stick them 05:18.020 --> 05:23.700 together, do whatever you feel like and build your borders. 05:23.720 --> 05:31.340 And again, my model, which I would be sharing as the code, is just one of these solutions. 05:31.670 --> 05:34.670 There is no absolute perfect solution. 05:35.090 --> 05:40.880 You can try at your own to create your own good solution for that. 05:41.420 --> 05:43.250 OK, so thanks. 05:44.960 --> 05:52.310 Please work on this particular model, building this particular problem and I will be sharing the solution 05:52.310 --> 05:54.290 in the next video for this one. 05:54.650 --> 05:55.070 Thank.