An empirical analysis of LADA diabetes case, control and variable importance

Latent Autoimmune Diabetes in Adults (LADA) is a condition, which is rarely recognised as a complex disease within its own right and remains under researched. Completely over-shadowed by Type 1 and Type 2 diabetes, LADA is the second most prevalent genre of diabetes after Type 2. This paper investigates conventional (clinical and socio-demographic) risk factors including Age, Gender, BMI (Body Mass Index), Cholesterol, Waist Size and Family History, with the motivation of determining their respective significant predictive power in the classification of LADA Diabetes. Such conventional factors are analysed and modelled using a set of supervised machine-learning algorithms including Support Vector Machines with Radial Basis Function Kernel (SVM), Random Forest (RF), K-Nearest Neighbour (KNN), Monotone Multi-Layer Perceptron Neural Network (MONMLP), Neural-net (NN) and Naïve Bayes (NB) Classifier, with the objective of correctly classifying LADA diabetes. Results elucidated from the analysis demonstrate that the predictive capacity of the learning models is significantly enhanced with the utilisation of Neuralnet classifier, achieving a classification accuracy of 85.51%, sensitivity of 84.09%, and specificity of 86.93%, alongside a precision of 86.93%, a recall of 84.53% and an F1 score of 85.71%, thereby outperforming the other studied individual models. Further analysis on the variable importance determined that the conventional variable Waist Size is the most significant variable when using the Neuralnet classifier with a 100% importance for LADA diabetes classification.