A Neural Ensemble Architecture with Pseudo-Random Input Sequence for Classifying LADA Diabetes

Latent Autoimmune Diabetes of Adult-onset (LADA) is a complex disease characterized by the attributes of both Type 1 and Type 2 diabetes, which makes distinguishing LADA challenging, particularly from Type 2 diabetes. Existing approaches for analyzing LADA utilize stand-alone statistical and/or machine learning algorithms, which have proven to be insignificant in capturing fine details pertinent to LADA. This study proposes a novel Neural Ensemble Model (NEM) to independently analyze LADA variables: first, conventional variables alone, then conventional variables in combination with five genetic variables above Genome Wide Association Studies (GWAS) levels, and finally conventional variables in combination with 11 genetic variables below GWAS levels. An ensemble carefully orchestrated with Radial Bias Function Feed Forward Neural Network (RBF-FFNN), a Multi-Layer Perceptron (MLP) with Feature Extraction and an MLP was developed to accurately classify LADA from Type 1 and Type 2 diabetes. The proposed NEM benefits from a pseudo-random input sequence from each genre of analysis to determine the most reliable identification of case, control, precision, recall, and variable importance, for maximum prediction accuracy. The proposed NEM achieves a greater prediction accuracy when utilizing a combination of conventional and genetic variables below GWAS levels, with 97.45% accurate identification of positive LADA diabetes.