Outbreak Cast-ML: Short-Term Disease Outbreak Analysis and Case Forecasting Using Machine Learning

Timely estimation of disease trends can help public-health teams plan testing, staffing, beds, medicines, and community communication. This paper presents OutbreakCast-ML, a reproducible machine-learning workflow for analysing reported outbreak data and producing short-horizon forecasts of daily cases and cumulative burden. The workflow combines data validation, daily incident-case derivation, exploratory visualization, time-ordered validation, and a comparison of a seasonalnaïve baseline, statistical time-series forecasting, and supervised regression models using lagged features. The supplied figures are included as illustrative historical visualizations of COVID-19 trends in India, including daily reported cases, cumulative cases, state-wise counts, and a Prophet-style forecast with trend components. These images are treated as reference graphics rather than evidence of current or independently validated model performance. No new numerical accuracy claim is made because the underlying dataset and model outputs were not provided for independent reproduction. The paper defines an evaluation protocol using MAE, RMSE, MAPE where mathematically appropriate, and predictioninterval coverage. It also discusses data quality, reporting delays, uncertainty, ethical communication, and the need for human review before operational use. The proposed contribution is a transparent end-to-end design that can be implemented and evaluated on public surveillance data and adapted to other infectious-disease settings.