ACADEMIC PRESENTATION / MACHINE LEARNING IN PUBLIC HEALTH
RESEARCH OVERVIEW

From Data to Diet:

Machine Learning Approaches for Predicting Household Nutrition Outcomes in Bangladesh

Nutrition is closely connected to a family’s economic condition.

CORE HOUSEHOLD PILLARS
PHYSIOLOGICAL VITALITY

Health

Physical well-being, growth, and disease resistance

HUMAN CAPITAL

Education

Nutritional awareness, literacy, and dietary diversity

ECONOMIC CAPACITY

Ability to Work

Productive labor capacity, earnings, and family income

MULTIDIMENSIONAL DETERMINANTS

WHAT AFFECTS A FAMILY'S NUTRITION?

INCOME
FOOD PRICES
FAMILY SIZE
TARGET OUTCOME

HOUSEHOLD NUTRITION

Vulnerability & Intake Quality
EDUCATION
EMPLOYMENT
LOCATION

Can we use household data and machine learning to predict which families are more likely to face food insecurity?

METHODOLOGICAL PIPELINE

FROM DATA → PREDICTION

HIES Household Income & Expenditure Survey
BDHS Bangladesh Demographic & Health Survey
01

HOUSEHOLD DATA

Income
Food expenditure
Family size
Education
Employment
Children
Rural / Urban
NEURAL ENGINE

MACHINE LEARNING

Logistic Regression
Decision Tree
Random Forest
Pattern recognition across multi-dimensional socio-economic vectors
02

PREDICTION

Food Secure Sufficient nutritional access & stability
Moderately Insecure Compromised diet quality & frequency
Severely Insecure High vulnerability; immediate support needed
IMPACT & ETHICAL CONSIDERATIONS

WHY DOES IT MATTER?

TARGETED ASSISTANCE
LIMITED RESOURCES
IDENTIFY VULNERABLE HOUSEHOLDS

need support, such as food assistance and nutrition programs

CORE SYSTEM DYNAMICS

THE POVERTY–NUTRITION CYCLE

BREAK THE CYCLE
LOW INCOME
POOR NUTRITION
POOR HEALTH
LOWER PRODUCTIVITY
CRITICAL CONSTRAINTS
CAUTION

BUT ML ISN'T PERFECT

Poor-quality data Incomplete surveys & missing attributes
Bias Underrepresented remote rural cohorts
Privacy Protecting household identifiable records
Prediction ≠ Causation Models correlate; systemic policy solves causes
SUMMARY & CONCLUSION

“So, overall, machine learning can help us understand nutrition problems and help the government make better decisions.”

THANK YOU

ARM Abir Hasan Computer Science & Machine Learning