dleutnant / drivendata-pumpitup

DrivenData-PumpItUp

This repository contains R code for the Pump it Up: Data Mining the Water Table competition on Driven Data.

The data is provided by Taarifa and the Tanzanian Ministry of Water. The goal is to predict whether a water pump is functional, functional but needs repairs or non functional.

I use H2O's random forest to get a score 0.821. I have uploaded my best (current) submission but not the data. Sign up at Driven Data to download the following files:

• SubmissionFormat.csv
• Test set values.csv
• Training set labels.csv
• Training set values.csv

Read the data and do some preprocessing

The first step is to read the data and set some values to missing (NA in R): See read-data.md.

Engineer features

The next step is to clean up the features (transform some, remove others) and possibly engineer some new features: See transform-data.md.

Predict status with a random forest

Use a random forest to predict the functionality status of pumps in the test set: See predict-data.md.

Update

Added a Makefile, which spins the R scripts to produces the md files. See how to Build a report based on an R script.