Towards Early Dengue Diagnosis in Bangladesh: A Non-Invasive Prediction Model Based on Symptoms and Local Trends

Dengue fever is a significant public health concern in Bangladesh, with a rising number of cases highlighting the need for early and accurate diagnosis. This study proposes a novel approach for early dengue diagnosis using a non-invasive predictive model based on external symptoms and local trends. Two datasets are prepared in the context of Bangladesh, one containing dengue cases with only external symptoms along with the month and district of affection. The other dataset contains statistical data on active dengue cases per month in Bangladesh from 2016 to 2023 and is used to capture seasonal trends. The number of cases is taken district-wise to determine the relation between dengue affection and active dengue cases in nearby areas. A two-step machine learning-based model is proposed, using machine learning-based time series forecasting models including the Long Short-Term Memory (LSTM) and Prophet models. The Prophet model performs better than the LSTM model, with lower MAE and RMSE values. A new dataset is created by merging the prediction of the first model with the external symptoms dataset's month and location attributes. Random Forest and XGBoost classifiers are used for case prediction on the second dataset and the merged dataset. The merged dataset has a top 82.3% accuracy, while the external symptoms dataset has a 92.5% accuracy. This approach aims to provide an efficient solution to improve early dengue diagnosis, ultimately facilitating timely intervention and resource allocation for effective disease management.