The Predictive Analytics Model is a desktop-native time series forecasting application built using Electron that performs client-side statistical analysis on marketing campaign data. It ingests CSV and Excel files through PapaParse and SheetJS, with automatic delimiter detection, date-format normalization, and heuristic column classification which works to fingerprint the source platform. 

The forecasting engine implements a total of five statistical models, including Holt-Winters triple exponential smoothing with grid search across smoothing parameters, an autoregressive model using Yule-Walker equations and Levinson-Durbin recursion, adaptive exponential smoothing on its tracking signal feedback, STL-inspired seasonal decomposition, and a damped trend model.

It uses walk-forward cross-validation, with a composite MAPE and R² scoring function, which is used to select the best-performing model for each dataset. The statistical pipeline also includes CUSUM change point detection, dual-method outlier detection combining both IQR fencing and Z-score thresholding, autocorrelation-based seasonality testing, as well as residual-based 95% prediction intervals that widen proportionally to the forecast horizon. 

The architecture follows Electron’s secure two-process model with context isolation enabled, node integration disabled, and a minimal preload IPC surface. 

The visualization layer uses Chart.js with time series plots featuring shaded confidence regions, trend decomposition charts, cumulative analysis, and sparkline widgets, with responsive theme adaptation. It runs client-side with no backend, ensuring data privacy and offline capability, and the app reports full diagnostics including MAPE, RMSE, R², volatility, and compound growth rate.