HYBRID VARIABLE AUTOMATIC ENCODER AND GRADIENT-BOOST DECISION TREE FOR ENHANCED DENGUE DISEASE PREDICTION AND DIAGNOSIS

Abstract

A Hybrid Variational Autoencoder and Gradient Boosted Decision Tree Model for Enhanced Prediction and Diagnosis of Dengue Disease describes an advanced deep-learning approach to improve Dengue disease prediction and classification. This system integrates a Variational Autoencoder (VAE) for efficient dimensionality reduction, enabling the extraction of key features from complex medical and environmental data. These compressed features are then used by a gradient-boosted Decision Tree (GBDT) model to accurately classify Dengue cases, leveraging gradient boosting with log-loss optimization and gradient descent for enhanced model performance. K-Fold Cross Validation is implemented to further prevent overfitting and ensure robustness, which divides the dataset into multiple subsets. This combined VAE-GBDT framework offers a powerful tool for early detection and precise diagnosis, supporting timely intervention and effective disease management in Dengue-affected regions.

Specification

Description:The present invention features an exemplary embodiment of the process for predicting and diagnosing Dengue disease using a hybrid deep learning model. The process begins with the collection of Dengue patient medical data (101), followed by data preprocessing to select key Dengue-related attributes (102). The preprocessed dataset, containing essential features, is then extracted using Variational Autoencoders (VAE) for dimensionality reduction (104), resulting in a reduced feature set (105). The dataset is subsequently split using K-Fold Cross Validation (106) to ensure robustness before training a Gradient Boosted Decision Tree (GBDT) model for classification (107). The accuracy and effectiveness of the VAE-GBDT model are then evaluated (108), followed by predictions that classify the cases as either Dengue Positive (110) or Dengue Negative (111). This methodology enables enhanced prediction and diagnosis of Dengue cases, ensuring more effective early detection and intervention. , Claims:1. A Design of an Intelligent Dengue Prediction and Diagnosis System (101) for automated detection, prediction, and classification of Dengue cases, comprising:
i. A data collection unit (102) that gathers medical data on Dengue patients, including symptoms, laboratory results, and relevant environmental factors, to provide comprehensive input for analysis (103); ii. A feature extraction module utilizing Variational Autoencoders (VAE) (104) for dimensionality reduction, creating a condensed dataset while retaining essential features for accurate predictions (105); iii. A classification unit powered by a Gradient Boosted Decision Tree (GBDT) model (106), which processes the reduced feature set to classify cases as either Dengue Positive (107) or Dengue Negative (108);
2. A Design of an Intelligent Dengue Prediction and Diagnosis System (101) as claimed in claim 1, wherein the VAE model applies a reparameterization trick (201) to facilitate backpropagation through sampling, reducing the data to a latent space while preserving key characteristics for prediction and analysis;
3. A Design of an Intelligent Dengue Prediction and Diagnosis System (101) as claimed in claims 1 and 2, wherein the system incorporates a decoder within the VAE that reconstructs the Dengue data from its latent representation (202), ensuring key features are preserved and enhancing the dataset for effective classification (203);
4. A Design of an Intelligent Dengue Prediction and Diagnosis System (101) as claimed in claim 3, wherein the GBDT model is trained using log-loss as the loss function (301) and optimized via gradient descent (302), with fine-tuning of hyperparameters based on Dengue-specific features to prevent overfitting and improve accuracy (303);
5. A Design of an Intelligent Dengue Prediction and Diagnosis System (101) as claimed in claim 4, wherein the system includes an evaluation phase (304) to assess the accuracy and effectiveness of the VAE-GBDT model, generate reliable classifications of Dengue cases, and update system performance metrics for ongoing improvements;and
6. A Design of an Intelligent Dengue Prediction and Diagnosis System (101) as claimed in claims 1 to 5, wherein the system integrates K-Fold Cross Validation (305) to ensure robustness by dividing the dataset into multiple subsets, preventing overfitting during model training and enhancing the model's generalizability.

Documents