Developing robust arsenic awareness prediction models using machine learning algorithms.

Area of Science:

• Environmental Health
• Public Health
• Socioeconomics

Background:

• Arsenic contamination poses a significant global health risk, particularly in rural areas.
• Sustainability of arsenic mitigation technologies depends heavily on community awareness.
• Socioeconomic dimensions of arsenic awareness remain underexplored, hindering effective prediction models.

Purpose of the Study:

• To evaluate arsenic awareness in rural Indian communities affected by arsenic.
• To identify socioeconomic, demographic, and sociobehavioral factors influencing arsenic awareness.
• To compare the predictive accuracy of machine learning algorithms for arsenic awareness.

Main Methods:

• A structured questionnaire collected data on socioeconomic, demographic, and sociobehavioral factors.
• Logistic regression and six machine learning algorithms (SVM, Kernel-SVM, DT, k-NN, NB, RF) were employed for prediction.
• Accuracy of prediction models was assessed to determine the best-performing algorithms.

Main Results:

• 63% of the surveyed population demonstrated arsenic awareness.
• Key predictors of awareness included caste, education, occupation, water collection practices, sanitation behavior, and social capital.
• Higher social network participation positively correlated with increased arsenic awareness.

Conclusions:

• Support Vector Machine (SVM) and Random Forests (RF) algorithms showed superior performance in predicting arsenic awareness.
• Vulnerable populations (lower caste, less educated, unemployed) require targeted arsenic mitigation strategies.
• Local institutions and NGOs are vital for enhancing arsenic awareness and outreach programs.