House price predictions

A Machine Learning Kaggle challenge that I made with my friend Felipe.

Overview

We tackled the Kaggle challenge House Prices - Advanced Regression Techniques in 5 days. It consists of predicting with Machine Learning Regression models the prices of suburbans houses in Ames, Iowa (USA).

The code is visible on Github.

About the team

Ironbuddies :

• Felipe de Ávila Granja Linkedin, Kaggle

• Lucie F

Dataset

Dataset is given by the Kaggle competition. It contains 43 categorical and 36 numerical variables describing the characteristics of residential homes in Ames, Iowa (USA).

The data description file can be found here

Main Steps

• Dataset exploration
• Selection of features (for numerical and categorical variables)
• Data cleaning
• Feature engineering
• Trainning + testing the model
• Improving Predictions
• Final testing

Techniques and tools

• Data visualization : correlation matrix, histograms, scatterplots,bars - [Matplotlib, Seaborn]
• Features tweeking :masked variables, one hot encoding, grouping and new feature creation (Neighborhood mean prices).
• Standardization : StandardScaler
• PCA (principal component analysis)
• Pycaret
• Random Forest regressor
• Hyperparameter tuning: gridsearch

Model

The model that we selected, after doing the Pycaret, was RandomForestRegressor.

Pycaret

This were the hyperparameters:

• n_estimators=100
• max_leaf_nodes=40
• max_depth=10

Final score