Using Feature Importance to Compute Key Determinants of Anaemia: A Case Study Based Approach

The research paper performs data analysis of primary health care data sets collected in a community setting, with the respondents being women in the 15-49 years age group in Assam, India. The size of the data sets is 650 records for a 99% confidence interval. There are five distinct data sets collected which are merged to get a consolidated data set. The data set includes diagnostic Complete blood count CBC data and social indices data along with nutritional and asset data. The social indices represent various variables such as region, marital status, occupation, source of drinking water, family size and number of rooms used for sleeping. Feature importance analysis and feature selection are done using random forests to compute the key social determinants of anemia along with a ranking of all 102 features of which ten most important are tabulated. The results obtained show that education, occupation and other biochemical variables are significant predictors of anemia. The analysis also shows that lower educational levels and occupation types are more prone to being anemic. These findings are in conformity with known determinants of anemia as identified in several studies available in the public domain.