[PYTHON] Moving vs Contract Renewal "Is the current rent reasonable?" Creating quantitative data to help make decisions

Introduction

I wonder if renters will have this kind of feeling as the time to renew their contracts approaches. "I'm not dissatisfied with my current house, but I wonder if it will be a little cheaper when I renew it ..." "I wonder if I should move if it's cheaper at the time of renewal" "Isn't the property rented higher than the rent market price ...?"

** "I want to check if the rent I'm paying is reasonable" ** This time, my goal is to quantitatively understand this.

Target

We will break down the analysis goals in detail.

** Who **: "The rental contractor (the person who lives)" ** When **: "When I'm worried about moving or renewing my contract" ** What **: "Is the rent you are paying reasonable?" ** If you know ** ** What you can do **: "You will be able to help you decide whether to move, renew your contract, or negotiate rent."

For this purpose, it is first necessary to know the rent market price of all properties.

Data preparation

Data is scraped from the rental property information service "SUUMO". Collection time: June 2020 Collection range: Tokyo's 23 wards Total number of data: 175,032

The collected features are as follows. (Refer to SUUMO) Based on these features, we will create a model that predicts rent. (The rent has been adjusted so that it is the sum of the management fees.)

Limited features

It is troublesome to enter a lot of information such as "how many floors" and "walk x minutes from △△ station on XX line" to grasp the market price of your rent. Following the idea of Occam's razor, we aim to create a model with "reasonable" accuracy with a small amount of features. We used lityt_GBM to check the importance of features. The accuracy is reasonably high and the learning cost is low, so it was very useful for the work of repeating "Which feature amount should I use?"

We have limited the final features to five.

We will make rent forecasts based on this feature.

What is the prediction accuracy?

I think, "Is that feature all right?" When rent forecasts were made using these five features, the following results were obtained.

I assumed that the rent price would vary greatly depending on the nearest station information, but it seemed that it was more related to which ward it belonged to than the nearest station. I was surprised here. In the prediction results, it was found that an error of approximately ± 12000 can be confirmed. We will continue to use a highly accurate random forest.

What is the rent for your own house?

We have completed the creation of a learning model that predicts rent based on features. Next, in order to confirm ** "Is the current rent appropriate?" **, it is necessary to create a new feature of the home, feed it with a learning model, and calculate the estimated amount (rent).

Refer to one property in Setagaya as a sample and feel like you are actually living. (I think I live on the 2nd floor) (Refer to [SUUMO](https://suumo.jp/jj/chintai/ichiran/FR301FC001/?ar=030&bs=040&ta=13&sc=13112&cb=0.0&ct=9999999&et=9999999&md=04&md=07&cn=9999999&mb=0&mt=9999999&shkr1] = 03 & shkr2 = 03 & shkr3 = 03 & shkr4 = 03 & fw2 = )）

Check the characteristics of the property.

This time, we will create a data frame using the input function.

Create a data frame. Calculation of prediction results using training data.

Consideration based on forecast results

The forecast result was cheaper than the actual rent. Whether to move, renew the contract, negotiate rent, What the landlord thinks here depends on each person and the surrounding environment. (I'm not actually living, so I'll refrain from commenting lol) However, it is information that can be taken into consideration in decision making.

** "Can you make it cheaper by the amount of management costs?" ** It may be okay to negotiate this much.

Finally

Was the goal ** "You will be able to help you decide whether to move, renew your contract, or negotiate rent." ** It became a model that contributed to this point to some extent. Even if the rent differs by several thousand yen, it will change to tens of thousands when converted to a year, so if you can create a model with even higher accuracy, it seems that you can contribute to more concrete decision-making.

After price negotiations, if the rent becomes cheaper even if it is 1000 yen, you can contract with Netflix with the expenditure unchanged. I can't underestimate ...

reference

If you want to reduce the rent of the property you live in now, I think the following YouTube will be helpful ~ https://www.youtube.com/watch?v=25DfwHHnKDg

Regarding SUUMO scraping, the following article was very helpful. https://qiita.com/haraso_1130/items/8ea9ba66f9d5f0fc2157