Speakers - 2025

Biography

Ms. Farasat Veisi is a researcher in molecular genetics at the Payame Noor University of Tehran Qeshm International Campus, Iran, specializing in the integration of artificial intelligence (AI) in biomedical research. Her work focuses on disease classification and prediction through machine learning and deep learning algorithms, aiming to enhance diagnostic accuracy and personalized treatment strategies. Her research explores AI-driven predictive modeling in healthcare. Farasat actively participates in international conferences, presenting her findings on the intersection of genetics, AI, and precision medicine. Her expertise spans bioinformatics, statistical modeling, and deep learning architectures such as CNNs, RNNs, and transformer models. Proficient in Python, R, and MATLAB, she applies advanced computational techniques to genomic data analysis and disease risk assessment. Her work is driven by a commitment to bridging genomics and AI, contributing to innovative solutions.

Abstract

Studying the spread of infectious diseases stands as a fundamental worldwide health priority. Traditional statistical models lack sufficient capacity to detect the intricate and dynamic spatial-temporal relationships that affect disease transmission patterns in various geographic areas. Open-access health data availability and deep learning technique development create an immediate need for data-driven frameworks that predict disease outbreaks precisely in both time and space. A new hybrid deep learning model merges Long Short-Term Memory (LSTM) networks and Graph Neural Networks (GNNs) to make predictions about infectious disease transmission across space and time. The model utilizes two datasets from Kaggle, which include the "Infectious Disease Prediction" dataset for California county-level disease case counts by disease type and gender across multiple years and the "Infectious Disease 2001–2014" dataset containing comprehensive outbreak data at the state level for different infectious diseases in the United States. The datasets received enhancement through the addition of historical weather data, including temperature and humidity measurements, together with demographic characteristics that included population density and urbanization data. The data instances contain three main components: spatial identifiers (state or county), weekly case count time series, and environmental and socioeconomic characteristic data. The model uses the weekly confirmed disease case counts from each region as its prediction output. The proposed methodology implements a spatio-temporal deep learning system that starts with GNN layers that use geographic distance and human mobility pattern adjacency matrices to model regional spatial connections. After spatial embedding processing, the LSTM layers analyze the sequential development of disease transmission. The model uses an 80-10-10 data split for training, validation, and testing while employing cross-validation to achieve both robustness and generalizability. The proposed hybrid model outperforms all baseline models, including ARIMA and standalone LSTM and random forest algorithms. The model achieved an RMSE of 7.41, which outperformed ARIMA at 12.35, and showed MAE at 5.83 compared to 9.72 for LSTM while obtaining R² at 0.89, which surpassed 0.73 from Random Forest. The model demonstrated successful identification of outbreak peaks and precise case surge predictions in dangerous areas during two-week periods. The addition of environmental and demographic variables resulted in an 11.4% improvement in the accuracy of predictions. The research presents an innovative approach for disease forecasting through the combination of LSTM and GNN architectural strengths. The proposed model provides a robust tool for early outbreak detection and resource allocation and strategic health intervention planning because it integrates spatial and temporal pattern analysis. Future improvements could be achieved by merging vaccination coverage data with public mobility patterns and international health data for better global application capabilities.