Predicting breast self-examination awareness in Sub-Saharan Africa using machine learning

Area of Science:

• Public Health
• Health Informatics
• Machine Learning in Healthcare

Background:

• Breast self-examination (BSE) offers significant cost reductions in healthcare.
• BSE improves health service accessibility and prevents infectious disease transmission in low- and middle-income countries.

Purpose of the Study:

• To identify key predictors of breast self-examination (BSE) practices.
• To evaluate machine learning models for predicting BSE behavior.

Main Methods:

• Utilized a large dataset (133,425 participants) from the Demographic and Health Survey.
• Applied data scaling, Recursive Feature Elimination, and ensemble techniques (Tomek Links, Random Over-Sampling) for model development.
• Evaluated model performance using metrics including AUC, accuracy, F1 score, recall, and precision.

Main Results:

• The Decision Tree model achieved the highest performance with 82% accuracy and 0.87 AUC.
• Key predictors identified include woman's age, smartphone availability, marital status, and healthcare access.
• The model's strength lies in capturing complex, non-linear relationships within the data.

Conclusions:

• Decision Tree models are effective for predicting breast self-examination behavior due to their ability to handle non-linear relationships.
• Recommendations include community leader engagement, mobile health initiatives, training health workers, and utilizing mass media for BSE awareness campaigns.