Preliminary analysis to gather some information for Machine Learning ALgorithm for the prediction of the house prices in Belgium.
We are working with a dataset created from scrapping the website inmoweb.be by looking for queries to get the prices of appartements and houses around Belgium. This dataset was cleaned applying the following criterias:
- Drop all the entirely empty rows
- Delete the blank spaces at the beginning and end of each string
- Fixing errors: Fixing variable hasFullyEquippedKitchen setting value '1' if the appartement has 'HYPER_EQUIPPED' or 'USA_HYPER_EQUIPPED' and '0' in the opposite case
- Dropping rows with price as NaN values
- Dropping duplicated values
- Deleting constant variable 'typeSale'
- Filling up empty values with np.NaN
We first visualiza the relationship between the vairables and the target (price) in a heatmap:
- The three variables with the highest correlation are:
| Variable | Correlation |
|---|---|
| area | 0.58 |
| BedroomsCount | 0.40 |
| hasSwimmingPool | 0.35 |
- The three variables with the highest correlation are:
| Variable | Correlation |
|---|---|
| hasSwimmingPool | 0.11 |
| hasGarden | 0.07 |
| isFurnished | 0.01 |
Then we explore the behavior of the house prices in relation to the area. The data was standarize the data taking the logarithm of the price and the square root of the area
We also looked at the average price per square meter in different regions of belgium (Brussels, Wallonia, Flanders) and in the whole Belgium.
We can see that in Wallonia we can find consierably less expensive houses and appartements than in Brussels or Flanders. Being Flanders the one with the most prices houses.
But at the same we can see that the mean price by square meter is bigger in Brussels.
