Application of ensemble models approach in anemia detection using images of the palpable palm

Abstract

Anemia is a public health issue with serious ramifications for human health globally. Anemia particularly affects pregnant women and children from 6 to 59 months old even though every individual is at risk. Anemia occurs when the Hb level is below its normal threshold or when the red blood cells are weakened or destroyed. To discover medical remedies on time, early detection or diagnosis of anemia assist patients to understand their condition.

The invasive approach for anemia detection is costive and time-consuming as compared to the non-invasive approach which is reliable and suitable for developing communities where medical resources and personnel are inadequate. This study uses palpable palm images (dataset) collected from 710 participants in selected hospitals in Ghana. The images were extracted, segmented and converted into RGB percentile to train, validate and tested the machine learning models. A hybrid model was developed with the application of ensemble learning models using the R programming language on the R Studio platform. Stacking, voting, boosting and bagging ensemble model techniques were used to build the hybrid models, the stacking ensemble model achieved an accuracy of 99.73 ​%. The study justifies that ensemble models are efficient for medical disease diagnosis or detection such as anemia.