Week 7- FINAL-Prediction of life quality
Until now we were developing our project as a solution to a regression problem,but in the last weeks, we came across a related work that was solved through classification and we have been thinking about how to get results if our problem is transformed into a classification problem. We tried to get results for both regression and classification models and we have seen that classification can be a good approach for this project.
For both approaches, we tried 3 methods to predict the scores:
• Support Vector Machines ( with 3 kernel functions: Linear, Sigmoid, RBF )
• Decision Trees
- Random Forest ( with bootstrap )
For classification, we labeled the life quality scores as a range 2, 4 and 10 class.
You can see the results below:
Unfortunately, we could not get the results we expected and there are a few reasons for this. For some cities, the geojson files that we used in the feature extraction, are quite large that we could not process, and we could not solve the “stop iteration” error we had in processing some files. So we were not able to use data fully (almost 1/3 of the data is not able to be used especially major cities and the small cities can keep more exceptions). In addition to this, the features were not stable enough and they caused the results to deviate considerably.
Although we could not achieve a successful result, we think that we have made the process analysis sufficiently. We are aware of what we need to achieve and what problems we need to solve, for a successful outcome. And we have got ideas for future of the process. Despite the time we have lost in the first weeks and the difficulties, we have experienced a project process that is as enjoyable as it is challenging.
