Dual level dengue diagnosis using lightweight multilayer perceptron with XAI in fog computing environment and rule based inference.

Over the last fifty years, arboviral infections have made an unparalleled contribution to worldwide disability and morbidity. Globalization, population growth, and unplanned urbanization are the main causes. Dengue is regarded as the most significant arboviral illness among them due to its prior dominance in growth. The dengue virus is mostly transmitted to humans by Aedes mosquitoes. The human body infected with dengue virus (DenV) will experience certain adverse impacts. To keep the disease under control, some of the preventative measures implemented by different countries need to be updated. Manual diagnosis is typically employed, and the accuracy of the diagnosis is assessed based on the experience of the healthcare professionals. Because there are so many patients during an outbreak, incompetence also happens. Remote monitoring and massive data storage are required. Though cloud computing is one of the solutions, it has a significant latency, despite its potential for remote monitoring and storage. Also, the diagnosis should be made as quickly as possible. The aforementioned issue has been resolved with fog computing, which significantly lowers latency and facilitates remote diagnosis. This study especially focuses on incorporating machine learning and deep learning techniques in the fog computing environment to leverage the overall diagnostic efficiency of dengue by promoting remote diagnosis and speedy treatment. A dual-level dengue diagnosis framework has been proposed in this study. Level-1 diagnosis is based on the symptoms of the patients, which are sent from the edge layer to the fog. Level-1 diagnosis is done in the fog to manage the storage and computation issues. An optimized and normalized lightweight MLP has been proposed along with preprocessing and feature reduction techniques in this study for the Level-1 Diagnosis in the fog computing environment. Pearson Correlation coefficient has been calculated between independent and target features to aid in feature reduction. Techniques like K-fold cross-validation, batch normalization, and grid search optimization have been used for increasing the efficiency. A variety of metrics have been computed to assess the effectiveness of the model. Since the suggested model is a "black box," explainable artificial intelligence (XAI) tools such as SHAP and LIME have been used to help explain its predictions. An exceptional accuracy of 92% is attained with the small dataset using the proposed model. The fog layer sends the list of probable cases to the edge layer. Also, a precision of 100% and an F1 score of 90% have been attained using the proposed model. The list of probable cases is sent from the fog layer to the edge layer, where Level-2 Diagnosis is carried out. Level-2 diagnosis is based on the serological test report of the suspected patients of the Level-1 diagnosis. Level-2 diagnosis is done at the edge using the rule-based inference method. This study incorporates dual-level diagnosis, which is not seen in recent studies. The majority of investigations end at Level 1. However, this study minimizes incorrect treatment and fatality rates by using dual-level diagnosis and assisting in confirmation of the disease.