Statistical analysis using data mining: A district-level analysis of malnutrition

Abstract

Children in India suffer from the highest level of undernourishment in the world, having serious effects on health. Bihar, in turn, has the highest incidence in India. This study was taken up to ascertain the effect of clusterization on a large dataset with respect to analyzes of the association of multitude of variables with malnutrition indices. Raw data were obtained using secondary data sources, especially government reports pertaining to malnutrition indices, demographic, social, nutritional, economic and medical factors causing malnutrition. In stage one the variables from un-clustered data were correlated with malnutrition indices (Stunting, wasting and underweight). Subsequently, the data was split using RapidMiner Studio (version-7.2.003) into three clusters. This segregation was done by the software using k-means and hierarchical agglomerative clustering. In the second phase, each of these clusters was again analyzed using the software and the correlation results were compared. Significant variation was observed in most of the correlations in the clustered and un-clustered datasets. This indicates the importance of clusterization in reaching the truth when a statistical analysis is carried out, as clusterization excludes/segregates the outliers/extremes of values. This has significant implications in policy making for malnutrition control, through identifying the most relevant variables/factors responsible.