Public repository for hosting the technical outputs of the CCI Nepal project.
This repository contains the code and documentation for a project uses Collective Crisis Intellegence (CCI) to help the Nepal Red Cross find the optimum Non-food-related-items (NFRI) for different households.
After a crisis strikes, people can be left without important supplies that support them to stay safe, healthy and comfortable. As part of coordinated efforts with other governmental and non-governmental organisations, the Nepal Red Cross Society (NRCS) provides Non-Food Related Items (NFRI) packages to affected communities.
A typical family NFRI package includes:
- Tarpaulin
- Blanket
- Sari, Male Dhoti
- Shouting cloth, printed cloth, plain cloth, teri cotten cloth
- Utensil set
- Water bucket
- Rope
These packages are often distributed in the aftermath of a crisis based on an Initial Rapid Assessment. During the project scoping phase, we interviewed NRCS team members to assess their needs and found that the team was interested in:
- Identifying a better way of knowing what NFRI items to distribute and when
- Exploring what new NFRI items community members are interested in
In line with these needs, the project uses new data collected from surveying two districts in Nepal - Sindhupalchok and Mahottari to:
- Measure the extent to which different households perceive existing items as essential (also referred to as item “essentialness” henceforth), and
- Collect and summarise suggestions on new NFRI items to include in packages in case of a flood crisis.
The analysis provided in this repository generates two outputs:
- A model that predicts item essentialness based on provided household information
- A stand alone piece of analysis showing what new items are suggested by households across different demographic features
Find out more about the project in our report Localising AI for crisis response.
Published on July 14, 2022
- Model workflow: Python code for train the models and then running on the test set.
- NFRI suggestion analysis: Analysis of the free text survey responses asking participants for suggestions of new NFRI items.
- Technical Specifications & XD Files: Comprehensive technical specifications for tool frontend and backend with proposed implementation.
This project uses new survey data collected from two districts in Nepal - Sindhupalchok and Mahottari. These districts were selected as they capture information from both the Hill and Plain regions in Nepal. Both districts are among the most flood affected districts of Nepal and have received NFRI from the NRC in the past. To design the survey we worked with members of the Red Cross and IFRC teams to develop the questions and ensure they were accessible.
3,265 responses were collected with a 50/50 split between male and female respondents, but otherwise using random sampling from households within the regions.
The survey was split into three sections
- Demographic information about the household: The first section asks the respondent a series of questions to understand the demographic characteristics of their household. In addition this section also asks for the households lat/long location and if they have previously received NFRI items.
- NFRI preferences of the household in a flood crisis: The second section asks the respondent to imagine a new flood crisis and asks them to state how important they see each NFRI item already distributed by the Red Cross.
- New NFRI items: The final section asks the respondent to suggest new NFRI items their household might need in a flood crisis that are not listed in section 2.
The first two sections are used for modelling where the first provides the features and the second provides our target (output) variables. The last section is not used for modelling but for a one-off piece of analysis for NRCS to help them understand new items that different groups might need.
The below table depicts the final features used by the model with their data type and a brief description.
| Column name | Description | Type |
|---|---|---|
| household_size | Count of total members in a household across all age groups. | int |
| percent_non_male | Percentage of total non male members in a household. | float |
| children_under_5 | A binary variable that represents if a household has any member less than age 5 or not. 1 representing yes and 0 representing no. | int |
| income_gen_ratio | Ratio of total income generating members in a household. | float |
| health_difficulty | A binary variable that represents if a household has any members with a heath difficulty. | int |
| sindupalchowk | A binary variable that represents if the district of the household is Sindupalchowk (represented as 1) or Mahottari (represented as 0), two districts that are present in the dataset. | int |
| household_material | A categorical variable representing the house material (Wooden pillar, RCC pillar, Bricks and stone, etc. ) | str |
Unfortunately the dataset is not publicly available due to it being a survey collected directly by the Red Cross and it containing sensitive data (more information here).
However, we have created a script that generates dummy data that allows you to test the running of model workflow. Follow the steps below to generate the dummy data and run the code.
- To run the models you will first need to setup the project. Follow the below two steps to do this:
$ git clone https://github.com/nestauk/cci_nepal
$ cd cci_nepal-
Run the command
make installto create the virtual environment and install dependencies -
Inside the project directory run
make inputs-pullto access the data from S3 (for those with access to the Nesta S3 account)
To note the project is setup using the Nesta Cookiecutter (guidelines on the Nesta Cookiecutter can be found here).
The below two options depend on if you have access to real survey data or need to generate dummy data to save and run the model.
Step 1: Save the file into inputs/data/
When saving your file make sure to save it in xlsx format.
Step 2: Update the file name in config
Navigate to cci_nepal/config and open the base.yaml file. In that file you will see the below file variable:
data:
file: "dummy_data"Change the value from dummy_data to the name of your file.
Run the below python file to create and save a dummy dataset that can be used for modelling. This is based on the questions used in our survey.
python cci_nepal/pipeline/dummy_data.pyRunning the dummy_data.py file saves a dummy version of the data you can use for modelling. The values are assigned randomly from the list of values for each column.
dummy_data.xlsx* saved in inputs/data.
*this is the default file used when you clone the repo. If you change the config file variable in option A you just need to remember to change it back to dummy_data if you want to re-run the script using your generated dummy data.
Perform the following steps to train and run the models:
Split the survey data into training / validation and test sets
python cci_nepal/pipeline/data_splitting_survey.pyThere are three files created from running the data_splitting.py file. These are saved in outputs/data/data_for_modelling and are listed below. These form the training, validation and test sets used for modelling.
train.csvval.csvtest.csv
Run the following modules to train, save and run the models.
python cci_nepal/pipeline/model_workflow/model_save.py
python cci_nepal/pipeline/model_workflow/model_test.pyThere are four files created from running the models and saved to outputs/data/test_evaluation_results:
shelter_test_predictions.csvwash_test_predictions.csvshelter_test_evaluation.csvwash_test_evaluation.csv
These contain the survey predictions and evaluation metrics for each shelter and wash/dignity NFRI items respectively. For the prediction files, the first set of columns will contain the feature names and the next set will contain the NFRI items with a 0 to 1 probability as to whether they are the item is predicted as essential.
The model contained within this repository were incorporated within wider technical architecture within the NFRI-Predict tool. This tool was evaluated using a high-fidelity prototype, and a technical specification is enclosed within this repository reports/technical-spec which outlines a proposed implementation of the wider tool architecture and how the model can be integrated. There is also an Adobe XD file included which shows the prototype tool.
The repository has the following main directories:
├── cci_nepal <- Packaged code (various modules, utilities and readme)
│ ├── analysis
│ │ ├── model_development <- Model tuning on the training set to find optimum models and parameters
│ │ ├── model_reporting <- Scripts to collect / report the results on the test set
│ │ ├── data_analysis <- Exploratory data analysis of the survey data
│ │ ├── free_text_analysis <- Analysis of the text questions of suggestions for new items
│ │ ...
│ ├── config <- Holds variables, feature names and parameters used in the codebase
│ ├── getters <- Functions for getting the data
│ ├── pipeline <- Holds scripts for all pipeline components
│ │ └── model_workflow <- Fits and saves the model using training data and runs it on test data
│ ├── utils <- Utility functions needed across different parts of the codebase
│ ...
├── inputs
│ └── data <- Holds original survey data (or dummy data)
│ ...
└── outputs
├── data
│ └── data_for_modelling <- Training, validation and test sets saved here
├── models <- Saved models after running model_workflow
├── reports
└── technical-spec <- Technical specification and XD files for tool implementation
...
This repository is maintained by the Data Analytics Practice team at Nesta. The CCI Nepal project is an ‘Accelerated Innovation Collaboration’ (AIC) which was delivered by Nesta’s Centre for Collective Intelligence Design (CCID) and Data Analytics Practice (DAP), UK HIH (information@ukhih.org), Open Lab, International Federation of Red Cross Red Crescent Societies - Solferino Academy, Cameroon Red Cross Society and Nepal Red Cross Society.
We welcome any feedback and questions. Please email dataanalytics@nesta.org.uk.