[WEEK I] Prediction Of Real Estate Price
Team Members: Ali Batuhan ÜNDAR, Enes Koçak, Muhammed İkbal Arslan
Oh look, it’s that time of year again with some more overdone project titles.
Jokes aside, I know that this have done about a googleplex times in the past but sadly finding a unique topic is harder than realizing project itself. That being said, while we progress through our little “science project” if we find any unique or close-to-unique ideas about said topic, we might just include them in the project. Even if it breaks it in someway.
What are you talking about?
Oh sorry, didn’t notice you were the “casul” type. So, in short what we are going to do;
- Find lots of information about a lot of houses. Like: how many room it has, where exactly it is in the city(or village, I am not judging), how big it is, how does it’s rooms look like(hard part)… The more, the merrier.
- Choose a machine learning model. Choosing to whether use images or not has great impact on this selection. But if we don’t even include image processing why are we even doing this, amirite?
- Jam the data into the model. Choose a program and flip the switch.
- Wait for ̶w̶a̶s̶h̶i̶n̶g̶ learning to be done.
- Do some prediction with a test data.
- Be disappointed with result. (do not mix colors and whites kids)
- Adjust model and parameters using the result. Flip again.
- Rinse and repeat ̶u̶n̶t̶i̶l̶ ̶y̶o̶u̶ ̶g̶e̶t̶ ̶t̶i̶r̶e̶d̶ ̶a̶n̶d̶ ̶c̶h̶o̶o̶s̶e̶ ̶a̶ ̶d̶i̶f̶f̶e̶r̶e̶n̶t̶ ̶t̶o̶p̶i̶c̶.̶
The idea is to give a somewhat decent price prediction based on the already known houses, like humans does (That should not be that difficult to understand considering you are one. Unless you are otherkin. Again, I am not judging). If you know nothing, there is no way you can make an accurate prediction. If you have bought few houses in the past then you can tell that whether you are scammed or not. Idea is the same.
While this may sound like
✌️ most terrific thing ever ✌️
this has already done by many students, scientists, engineers, astranouts and my 80 years old neighbor. You can say, this is more of an example than a real project. That’s why there is a lot of mixed feeling happening here.
Oh come on, you are exaggerating.
Maybe a little. I like to exaggerate. But of course, the idea is not new. People have been researching this topic for few years now. Here are some articles as an example:
- https://arxiv.org/abs/1707.05489: This article talks about how interior and exterior photos of houses impact the price prediction made by computer. Might be useful.
- https://arxiv.org/abs/1808.02547: This one is about how the surrounding neighboorhood affects the price of a real estate. Not exactly the same thing but helpful.
- https://arxiv.org/abs/1802.08238: This one is about which parameters affects the house prices and how much they affect. Very helpful when doing fine tuning.
- https://arxiv.org/abs/1611.09180: This one is pretty cool. It only uses location and image data to make prediction. Meaning it make predictions mostly based on images. And they claim their method is pretty accurate too. Interesting…
Ok then. How is it going so far?
Nothing much so far. That’s very nice of you to ask, surprisingly so. We have written few scripts to s̶t̶e̶a̶l scrape data from few real estate agencies. So far we have collected about 1000 house data(well, links). A few more adjustments and we will have lots of real world data.
Except, photos. So idea is to use photos of houses to improve our predictions. That makes data gathering so much more difficult. Most of websites does not tag their photos so we have no idea what room the given photo is from. That is the first problem we encountered. Now there is three ways to solve this;
- Seperate photos by hand (so much time wasted)
- Use the f̶o̶r̶c̶e machine learning to seperate images(and if we do this successfully, is there even need for finishing the rest of the project. That is a topic on it’s own and somewhat unique.)
- c̶r̶y̶ Use a read-to-use dataset.
So overall progress is, well, nonexistent. However, we have somewhat clear path and it shouldn’t be that difficult to meet deadline with our topic. The problem is giving it a unique flavor. But you can be sure of that we are going to try to find it.
Thanks for reading and, please, take this:
