A new design of diabetes detection and glucose level prediction using moth flame-based crow search deep learning. (August 2022)

• Record Type: Journal Article
• Authors: Naveena, Somasundaram; Bharathi, Ayyasamy
• Is Part Of: Biomedical signal processing and control. Volume 77(2022)
• Publication Date: 2022-08
• Subjects: Diabetes detection -- Glucose level prediction -- Hybrid meta-heuristic-based CNN -- Modified fuzzy classifier -- Enhanced recurrent neural network -- Moth flame-based crow search algorithm; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.103748
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Perform diabetes detection and glucose level detection using classifiers by gathering the data from PIMA and UCI datasets. Accomplish the deep feature extraction using CNN by optimizing its parameters for maximizing the correlation of attributes. Detect diabetes using the modified Fuzzy classifier by optimizing the parameters of Fuzzy for increasing the accuracy. Predict the glucose level by the enhanced RNN that minimizes the computational time and minimizes the RMSE and MASE. Implement an MF-CSA for handling the multi-objective optimization problems and it improves diabetes detection. Abstract: The main intent of this work is to implement an intelligent method for the diabetes detection and blood glucose level prediction. Initially, the data's are taken from the significant benchmark datasets known as the PIMA and the UCI dataset. It is subjected to the deep feature extraction using hybrid meta -heuristic-based Convolutional Neural Network (CNN) with two max pooling and two convolutional layers. A new novel algorithm is developed named MF-CSA that can handle the multi-objective optimization problems and it improves the deep feature extraction, diabetes detection, and glucose level prediction process. As the main contribution, the optimal feature selection by the combination of deep learning models like CNN is performed with Moth-Flame Optimization (MFO) and Crow Search Algorithm (CSA) that is called as the Moth Flame-based Crow Search Algorithm (MF-CSA). The hybrid MF-CSA is used to optimize the count of hidden neurons in two convolutional layers in CNN to attain minimum correlation between the features to avoid redundant information. With these features, the diabetes detection is performed by modified fuzzy classifier with membership optimization. This provides the output as high glucose or low glucose level. Once the levels are classified, its range is predicted by the enhanced Recurrent Neural Network (RNN) based on the suggested MF-CSA. The comparison of the proposed model with existing optimization as well as classification algorithms is done to prove the superiority of the implemented model. Thus, the simulation outcomes recommend that the suggested model can enhance glucose prediction performance effectively when compared over the existing approaches.