user@devops:~$ cat README.md
Time Series Forecasting
# Description
Complete time series forecasting project with two complementary approaches: Facebook Prophet (additive statistical model with multiple seasonality) and LSTM (recurrent neural network with long-term memory in PyTorch). Includes seasonal decomposition (trend/seasonality/residual), Augmented Dickey-Fuller stationarity test, 30-day forecast with 95% confidence intervals, and quantitative metric comparison (MAE, RMSE, MAPE). Results show Prophet achieves 1.87% MAPE on clear seasonal data, while LSTM reaches 15.96% on a small training subset.
# Key features
$ Classic time series decomposition (trend, seasonal, residual)
$ Augmented Dickey-Fuller (ADF) stationarity test
$ Facebook Prophet with multiple seasonality (daily/weekly/yearly/monthly)
$ LSTM in PyTorch with 2 layers, Huber Loss, and early stopping
$ 95% confidence intervals with Prophet
$ Quantitative comparison across 3 metrics (MAE, RMSE, MAPE)
$ Visualizations: EDA, decomposition, components, forecast, training curves
# Gallery
# Technologies used
