vadimbs/tanzanian-water-pumps-status-analysis

This analysis was done to better understand and find ways to improve water pumps operation and consequently the quality of life in Tanzania. # Tanzanian water pumps status analysis This analysis was done to better understand and find ways to improve water pumps operation and consequently quality of life in Tanzania. Model was built for prediction water pumps status on data from Taarifa and the Tanzanian Ministry of Water . ## Contents * EDA.ipynb - preprocess data (exploratory data analysis) * visualisation.ipynb - all visualisations done through EDA * student.ipynb - main file with built model * import_libraries.py - helper file for import necessary libraries * helper.py - helper functions * data/ * downloaded/ - folder with data from drivendata.org * train_data.csv - train features * train_labels.csv - labels * test_data.csv - test features * SubmissionFormat.csv - test labels (which sould be predicted) * preprocessed/ * data.pickle - selected features(filled na) and labels in one dataset * test_comb_cats.pickle - test part of data.pickle with combined * train_comb_cats.pickle * train_comb_cats_cleaned.pickle * predicted_test_labels.csv * test_for_SM.pickle * train_for_SM.pickle * model/ - turns out that models are to big for uploading to github * model1.pickle - 1st iteration of models fitting * model2.pickle - 2nd iteration of models fitting * model3.pickle - 3rd iteration of models fitting * model4.pickle - 4th iteration of models fitting ## Conclusions We fitted several classification models and get the best result with RandomForestClassifier() and XGBClassifier(). Both behave well after tuning hyperparameters. #### Submission with only `RandomForestClassifier()` model (accuracy score: 0.802) #### Submission with ensemble of `RandomForestClassifier()`, `XGBClassifier()` and `DecisionTreeClassifier()` using `VotingClassifier()` (accuracy score: 0.806) #### In this particular case ensemble method did worse than the random forest classifier with accuracy score 0.7996 and 0.8055 respectively