USAID's Intelligent Forecasting Challenge: Model Future Contraceptive Use

Build and validate your models on Zindi. Make your final submission directly to USAID.
The goal of the Intelligent Forecasting Competition is to predict consumption (stock_distributed in the Primary Data Dictionary) for 11 contraceptives across 156 health service delivery sites in the public sector health system in Côte d’Ivoire. The predictions should be made monthly for three months: October 2019, November 2019, and December 2019, using the dataset provided. The data field that grantees will be predicting is stock_distributed. The Zindi leaderboard scores predictions for three months: July 2019, August 2019, and September 2019.
Zindi competitors will have access to 42 months, January 2016 through June 2019, of data. The featured dataset is an extract from the electronic logistics management information system (eLMIS) used in Côte d’Ivoire to manage contraceptives and other health products across the country’s public health system.
Files available for download
Train.csv - you will use this data to train your model on
SampleSubmission.csv - is an example of what your submission file should look like. The order of the rows does not matter, but the names of the IDs must be correct.
monthly_cases.csv - monthly information about distribution of contraceptives in from January to June 2019
annual_cases.csv - annual information about distribution of contraceptives per health center for 3 years
data_dictionary.csv - variable defintions for each csv file
product.csv - defintion of each contraceptive
Recommended_Supplementary_Data_Sources.docx - a list of extra data recommended by USAID
service_delivery_site_data.csv - information about delivery sites, including the latitude and longitude
Table_of_Contents.xlsx - description of the files available
contraceptive_case_data_annual.csv - annual information about distribution of contraceptives per health center for 3 years
Data quality in data sets such as these is a significant issue. For example, not all sites report all data each month. In addition, there are cases in the contraceptive_logistics_data.csv dataset where the Côte d’Ivoire eLMIS system records a '0' value that may in fact represent a "non-reported" value or a value entered by a user for convenience. In such an instance, these ‘0’ values do not reflect actual observed values.
These ‘invalid 0 values’ may be identified, for example, when:
A) All fields (possibly with the exception of stock_ordered) are recorded as '0' for a given contraceptive product, month and service delivery site; or
B) Stock_distributed for a contraceptive product at a service delivery site is reported in a given month as '0', even though stock was available (stock_initial was above 0) and stock_distributed is historically well above 0.
These apparently invalid '0' values have not been cleaned from the primary dataset provided.