Optimisation of deep learning models using flock algorithm in improving feature modelling and diabetes classification

Diabetic illness is a widespread health issue that affects people of both genders. The diabetic disease creates several diseases, such as heart problems, stroke, etc. Therefore, patients are doing a self-assessment process to identify the diabetic disease in the earlier stage. However, the self-assessment process fails to detect most cases, leading to increased complexity. The prediction model requires robust training model due to the insufficient data assimilation that leads to the generalization problem and minimize the prediction issues. Another important problem is imbalanced data which occurred due to the uneven distribution of diabetic classes (non-diabetic cases and diabetic case). The imbalance problem having influence on the learning ability and accuracy; Therefore, an automatic detection system is created to identify diabetic disease at an earlier stage. However, the existing diabetic prediction models fail to address the data assimilation, imbalance data, information loss, and accuracy issues. Therefore, this research proposed a method called the flock-optimized assimilated deep learning model (FOADLM) to improve diabetic detection efficiency. The flock optimization approach maximizes model convergences and minimize training time which leads to improve the diabetic detection efficiency. The PIMA Indian Dataset, Azure Dataset, and OhioT1DM Dataset information are used to evaluate the accuracy of FOADLM. Initially, the Gaussian Filtering is used to process the data by using the kernel value to filter out irrelevant information. Then neural convolution model is integrated with the flock optimization technique to determine the features. The effective utilization of optimization techniques recognizes the relationship between the data, minimize the information loss, and improve the overall data accuracy. Then data sampling techniques is applied by combine with the flock optimization in Convolution Neural Networks (CNN)’s first layer achieves the system flexibility. During the analysis, pre- dominant and latent features are utilized in the classification that minimize the computation complexity and improves the overall recognition accuracy. The Long Short Term Memory Neural Networks (LSTM) network analyses the discovered features to identify diabetes effectively. Thus, the system achieves 99.23% accuracy and minimum error rate (0.1) as findings, surpassing the previous works and various diabetic detection systems.

Text FSKTM 2024 19 - Full Text.pdf Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (14MB) Text FSKTM 2024 19.pdf Restricted to Repository staff only Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (15MB)

Actions (login required)

View Item