In this project we predict house sale prices using housing data for the city of Ames, Iowa, United States from 2006 to 2010.
transform_features() - Within this function we remove the non-numerical and other features that are not meaningful for our model. Besides that we remove all features that have more than 25% missing values. Finally we either delete the rows (for non continuous features) or replace the missing values for the mean()
select_features() - After identifying the features which are highly correlated, we remove those of the pairs, which have less impact on the model. Within this funciton we also re-scale the features and remove those with very low variance. train_and_test() - Here we train and test our model with the remaining features. We perform the process by splitting the set in 50/50 (train/test) and also try different cross-validation combinations.
Results show that, for this dataset, increasing the number of datasets in the k-fold cross validation leads to always smaller RMSEs