A Machine Learning and Deep Learning Framework for Spatiotemporal Prediction of Dengue Incidence

Abstract

This study offers a comprehensive model for accurate dengue forecasting based on a combination of machine and deep learning. Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), K-Nearest Neighbor (KNN), Naïve Bayes (NB), Support Vector Machine (SVM), Recurrent Neural Network (RNN), Feedforward Neural Network (FNN),Dense Neural Network (DNN), Long Short-Term Memory (LSTM) — are fully covered by the proposed system. The dataset, downloaded from ScienceDirect database, has 1,523 de-identified patient records with 19 clinical and hematologic attributes as blood values demographics, laboratory tests, and confirmed dengue results. The data pre-processing step to ensure good quality input, involved encoding of categorical attributes, deletion of duplicates, balancing, scaling, and exploratory data analysis with statistical summaries and visualizations. Testing and training data served to divide the data into training (80%) and testing (20%) to minimize overfitting and to ensure fair evaluation of performance. The performance of classification was quantified in terms of accuracy, precision, recall, F1- score, and ROC-AUC, and each model was trained and tested independently. The RF, second only to the SVM and the deep learning-based models, also proved its effectiveness in dengue classification, as it had the highest prediction accuracy of 77% among all the models. On the other hand, KNN resulted in poor results at the same time logistic Regression and Naïve Bayes resulted in fair results. The effectiveness of computational intelligence on early dengue detection, early diagnosis, and public health surveillance is shown to be efficient here. In future, data imbalance should be addressed for superior predictive healthcare analytics, interpretability should be enriched, and more clinical signs considered.