The target is a derived ‘wealth index’ based on responses in various household surveys. It is on a scale from 0 to 1, with 1 denoting higher wealth. This measure is based on multiple factors such as asset ownership, and gives a single measure across the different surveys and countries that we can use to compare data points in a consistent way. Each row in the dataset corresponds to a cluster of survey responses, and so the Target is a measure of the average wealth for a given cluster of respondents.

For the solution to be applicable elsewhere, the inputs need to be derived from datasets with appropriate coverage. All of the datasets used cover the whole of Africa, meaning that any solution based on these can in theory be used to predict the wealth index for any country. Layers sampled include measures of human settlement, land cover and nighttime light emissions, as well as two distance measures that often relate to economic success: distance to the country’s capital and distance to the ocean. For many of these measures, the value given is the average for a 5km radius around the cluster center.

If, when you click to download the starter notebook it takes you to another page, ctrl-S and it will save to your downloads folder.

Variable definitions

Global Human Settlement Layer. Based on data from the GHSL project.

• 'ghsl_water_surface' - the fraction of land within 5km of the cluster that is classified as water surface
• 'ghsl_built_pre_1975' - the fraction of land within 5km of the cluster that is classified as built-up before 1975
• 'ghsl_built_1975_to_1990'
• 'ghsl_built_1990_to_2000',
• 'ghsl_built_2000_to_2014',
• 'ghsl_not_built_up' - land that was never built up
• 'ghsl_pop_density' - population density for the surrounding area (5km radius)

Landcover: based on the Copernicus Global Land Cover Layers (https://developers.google.com/earth-engine/datasets/catalog/COPERNICUS_Landcover_100m_Proba-V-C3_Global)

• 'Landcover_crops_fraction' - the fraction of land within 5km of the cluster that is classified as cropland
• 'landcover_urban_fraction' - the fraction of land within 5km of the cluster that is classified as urban
• 'Landcover_water_permanent_10km_fraction' - the fraction of land within 10km of the cluster that is classified as permanent water
• 'Landcover_water_seasonal_10km_fraction' - the fraction of land within 10km of the cluster that is classified as seasonal water
• 'Nighttime_lights' - a classic indicator of economic activity
• 'Dist_to_capital' - distance to the countries capital
• 'Dist_to_shoreline - distance to the nearest ocean shoreline
• ‘urban_or_rural’: Is the cluster in an urban (‘U’) or rural (‘R’) setting

Files available for download:

• Train.csv - contains the target. This is the dataset that you will use to train your model.
• Test.csv- resembles Train.csv but without the ‘Target’ column. This is the dataset on which you will apply your mode.
• SampleSubmission.csv - shows the submission format for this competition, with the ‘ID’ column mirroring that of Test.csv and the ‘Target’ column containing your predictions. The order of the rows does not matter, but the names of the ID must be correct.
• Starter_Notebook.ipynb - this is a starter notebook that will help you look at the data and make your first submission.