Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Predicting household socioeconomic position in Mozambique using satellite and household imagery

Carles Milà, Teodimiro Matsena, Edgar Jamisse, Jovito Nunes, Quique Bassat, Paula Petrone, Elisa Sicuri, Charfudin Sacoor, Cathryn Tonne

TL;DR

The results show how ground-based household photographs allow to zoom in from an area-level to an individual household prediction while minimizing the data collection effort by using explainable machine learning.

Abstract

Many studies have predicted SocioEconomic Position (SEP) for aggregated spatial units such as villages using satellite data, but SEP prediction at the household level and other sources of imagery have not been yet explored. We assembled a dataset of 975 households in a semi-rural district in southern Mozambique, consisting of self-reported asset, expenditure, and income SEP data, as well as multimodal imagery including satellite images and a ground-based photograph survey of 11 household elements. We fine-tuned a convolutional neural network to extract feature vectors from the images, which we then used in regression analyzes to model household SEP using different sets of image types. The best prediction performance was found when modeling asset-based SEP using random forest models with all image types, while the performance for expenditure- and income-based SEP was lower. Using SHAP, we observed clear differences between the images with the largest positive and negative effects, as well as identified the most relevant household elements in the predictions. Finally, we fitted an additional reduced model using only the identified relevant household elements, which had an only slightly lower performance compared to models using all images. Our results show how ground-based household photographs allow to zoom in from an area-level to an individual household prediction while minimizing the data collection effort by using explainable machine learning. The developed workflow can be potentially integrated into routine household surveys, where the collected household imagery could be used for other purposes, such as refined asset characterization and environmental exposure assessment.

Predicting household socioeconomic position in Mozambique using satellite and household imagery

TL;DR

The results show how ground-based household photographs allow to zoom in from an area-level to an individual household prediction while minimizing the data collection effort by using explainable machine learning.

Abstract

Many studies have predicted SocioEconomic Position (SEP) for aggregated spatial units such as villages using satellite data, but SEP prediction at the household level and other sources of imagery have not been yet explored. We assembled a dataset of 975 households in a semi-rural district in southern Mozambique, consisting of self-reported asset, expenditure, and income SEP data, as well as multimodal imagery including satellite images and a ground-based photograph survey of 11 household elements. We fine-tuned a convolutional neural network to extract feature vectors from the images, which we then used in regression analyzes to model household SEP using different sets of image types. The best prediction performance was found when modeling asset-based SEP using random forest models with all image types, while the performance for expenditure- and income-based SEP was lower. Using SHAP, we observed clear differences between the images with the largest positive and negative effects, as well as identified the most relevant household elements in the predictions. Finally, we fitted an additional reduced model using only the identified relevant household elements, which had an only slightly lower performance compared to models using all images. Our results show how ground-based household photographs allow to zoom in from an area-level to an individual household prediction while minimizing the data collection effort by using explainable machine learning. The developed workflow can be potentially integrated into routine household surveys, where the collected household imagery could be used for other purposes, such as refined asset characterization and environmental exposure assessment.

Paper Structure

This paper contains 23 sections, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Materials and methods
  3. Study area and population
  4. Sampling procedure
  5. Data collection
  6. Primary data collection
  7. Secondary data collection
  8. Satellite imagery
  9. Final dataset and missing data
  10. Analysis
  11. Construction of SEP measures, exploratory analyses, and data split
  12. Extraction of feature vectors from the images
  13. Regression analyses
  14. Explaining and optimizing models
  15. Implementation and code availability
  16. ...and 8 more sections

Figures (6)

  • Figure 1: Workflow of the data collection and analysis in the SEP study.
  • Figure 2: Examples of all image types collected in the study.
  • Figure 3: Graphical representation of the transformed VGG16 CNN architecture used in the study for feature vector extraction.
  • Figure 4: Exploratory analysis of the SEP measures: assets (MCA first dimension), expenditure (metical), and income (metical). The diagonal panels show the histograms of the SEP measures and their descriptive statistics (SD: Standard Deviation; IQR: InterQuartile Range), while the lower and upper off-diagonal panels show the bivariate scatterplots and the Pearson ($r$) and Spearman ($\rho$) correlation coefficients, respectively.
  • Figure 5: Light source images with the largest positive (top row) and negative (bottom row) SHAP values in random forest complete models. Images were selected according to the top and bottom average SHAP value ranks across the three SEP measures.
  • ...and 1 more figures