gersongerardcruz / regression_analysis_on_house_pricing

This repository contains a jupyter notebook focusing on Kaggle's House Pricing Dataset. Exploratory Data Analysis, Data Preprocessing, and Modelling was performed using Pandas, Numpy, Seaborn, and Scikit-learn

Regression Analysis on House Pricing

This repository contains a jupyter notebook focusing on Kaggle's House Pricing Dataset. Exploratory Data Analysis, Data Preprocessing, and Modelling was performed using Pandas, Numpy, Seaborn, and Scikit-learn.

Dataset

The dataset consists of a train and test set gotten from Kaggle's Datasets at this link: https://www.kaggle.com/c/house-prices-advanced-regression-techniques/data

For more information about the datasets, check out the data_description.txt file!

Project Features

Exploratory Data Analysis

• Pandas functions were used to gather information about the statistical summary of the dataset, missing data points, and quantitative and qualitative features
• Seaborn and Matplotlib functions were used to visualize the relationship between the target variable Sale Price to other quantitative features.

Data Cleaning and Preprocessing

• Imputers were utilized to replaced certain features with simple mean, median, and mode
• Lambda functions were used to replaced multiple NaN rows with the appropriate None or NA categorical data
• Category encoders were used to replace categorical features. Binary and Pandas' Get_Dummies were the main encoders dependent on the number of unique values per column.

Data Modelling

• Scikit-learn's StandardScaler was used to scale the data to avoid outliers affecting the quality of the model.
• Numpy's logarithmic transformation was used to make the target data follow a normal distribution, as required in any regression analysis
• Skikit-learn's LinearRegression, GradientBoostingRegressor, and HistGradientBoostingRegressor the main models in this notebook.
• Hyperparameter tuning was performed via the selection of the best learning rate for the GradientBoosting Regressors.
• The main metric in this notebook was the Root Mean Squared Error (RMSE), which after logarithmic transformation, decreased from an average of 30000 to 0.11

Recommendations

For future iterations of this project, the following improvements can be done:

• More use of hyperparameter tuning to select best features for the models and the use of more models in doing the training and validation
• For Data Cleaning and Preprocessing, the results of directly removing uncorrelated variables can be explored
• The use of model ensembling to further decrease the root mean squared error