An Efficient Cat-Boost Classifier optimized with Woodpecker Mating Algorithm to predict and diagnose Type 2 Diabetes

Abstract

Abstract Type 2 Diabetes Mellitus (T2DM) is a chronic condition marked by the body's incapability to break down glucose. In the current scenario, screening tests for diabetes are developed with the help of multivariate regression methods. The growing capacity of information is collected automatically to provide a chance for developing accurate and more complex prediction models that could be updated continuously using machine learning techniques. In this manuscript, an efficient Cat-Boost classifier optimized with the Woodpecker Mating Algorithm is proposed to predict and diagnose type 2 diabetes mellitus. Initially, data are taken from the PIMA Indian Diabetes Dataset (PIDD), and then they are transferred to the preprocessing segment. Here, the pre-processing process is done with the help of Savitzky-Golay Denoising method, which removes the noise in the input data. The preprocessed data is transferred into the Cat-Boost Classifier (CBC) for classifying the diabetic and non-diabetic. Finally, the hyperparameter of the Cat-Boost Classifier (CBC) is tuned with the Woodpecker Mating Algorithm (WMA) which helps to gain better classification of diabetic and non-diabetic. The proposed method is executed in Python and its performance is examined with evaluation metrics like accuracy, f-measure, specificity, sensitivity, precision, and error rate. The proposed method attains higher accuracy (99.57%, 97.28%, 96.87%, and 96.34%), higher sensitivity (89.94%, 88.90%, 85.34%, and 90.65%), and a lower error rate (17.8%, 20.52%, 17.03% and 15.55%) compared with existing methods like predicting the onset of T2DM using a wide and deep feed-forward neural network with electronic health records (DFFNN-PD-T2D), type 2 diabetes mellitus prediction using a deep neural network classifier (DNN-PD-T2D), type 2 diabetes data classification using stacked auto encoders in DNN (SAEDNN-PD-T2D) and deep learning method based on convolutional LSTM for detecting type 2 diabetes (CLSTM-PD-T2D).