Abstract

The continuous occurrence of infectious disease outbreaks proves the necessity for precise and immediate forecasting methods to tackle global health challenges. The traditional surveillance systems use past data for analysis, but they do not have predictive features, which restrict their capacity for early response preparation. This study develops a deep learning system that combines temporal and demographic data to make predictions about infectious disease outbreaks. The integration of time series and demographic data provides the model with improved predictive capabilities for public health monitoring. The study analyzes two publicly available datasets from Kaggle, which are titled "Infectious Disease Cases Data" and "Infectious Disease Prediction." The main target variable measures confirmed infectious disease cases that occur daily or weekly. The analysis depends on input data, which includes past disease cases combined with date information along with seasonal indicators together with demographic elements like population density, age group distribution, gender ratio, and geographical region. External environmental data about temperature and humidity can be integrated as optional features to enhance model context understanding. The proposed approach uses a hybrid deep learning framework that includes Long Short-Term Memory (LSTM) networks and an attention mechanism. The temporal input stream runs through LSTM layers to extract time-dependent patterns, which combine with a dense layer to process static demographic features. A subsequent attention mechanism processes the temporal data to determine which time steps hold the most significance. The predictive model combines output from both branches to produce its final predictions. The assessment of model performance compares against three traditional forecasting methods: ARIMA, Random Forest, and XGBoost. The experimental findings show that the proposed model demonstrates strong predictive performance along with universal applicability. The model demonstrates performance metrics of RMSE at 13.42 and MAE at 9.81 and an R² score of 0.91 during multiple cross-validation experiments. The model achieves a binary classification accuracy rate of 94.7% when it is used for outbreak risk prediction through a defined threshold classification. The attention visualization results show that the most influential factors for prediction include recent seven-day case values and regional population density, thus demonstrating the model's practical and interpretive value. The study presents an original solution to predict infectious disease outbreaks by merging temporal patterns and demographic variables through a clear deep learning framework. LSTM integration with attention mechanisms produces precise forecasting outcomes and improves outbreak driver understanding. This method shows strong potential for deployment as an early warning system that enables data-driven public health decisions at the right time.