ErinNC/Tanzania-Water-Pumps

Predictive classification project for working status of water pumps in Tanzania using a variety of models # Tanzania Water Pump Classification ## Description: A large portion of Tanzania's population lack access to potable water, so it is imperative to know the functional status of pre-existing water pumps. Using data provided by Taarifa and Tanzania Ministry of Water, we will predict whether a pump is `'functional'`, `'functional needing repair'`, or `'non-functional'` using a variety of machine learning models. The underlying assumption is that the Tanzanian government lacks resources and funds to inspect every single water pump. Our machine learning model will provide predictions for the pumps in the most dire need of repair. We will use **decision tree**, **random forest**, and **gradient boosting classifiers** and compare their results using classification accuracy as our evaluation metric. ## Necessary Packages: - matplotlib - numpy - pandas - pandas_profiling - scikit-learn - scipy - seaborn - sklearn - xgboost ## Contents: In this repository, you will find: - An **exploratory data analysis notebook**, which we will use to create a data wrangling function for our machine learning notebook - A **machine learning notebook**, in which we will prepare and split our data using **pipelines**, train our models, compare evaluation metrics, and tune our models using several cross-validation methods. - An html file containing our **Pandas Profiling Report** ## Credits: This predictive modeling challenge comes from DrivenData, an organization that helps non-profits organizations by hosting data science competitions for social impact. The competition has open licensing: "The data is available for use outside of DrivenData." This project was also part of BloomTech's Data Science curriculum. ## Next Steps: One way of improving our model in the future could be to change how we handle high cardinality categorical features. Rather than just dropping these columns, we could reduce the cardinality of each feature by aggregating the categories, using an "other" field. We cou …