Overview
Developed a time-series forecasting model using Facebook Prophet to predict future retail sales from historical transaction data. The system performs trend analysis, seasonality detection, and generates a 30-day sales forecast with confidence intervals.
Technical Highlights
Data preprocessing & datetime normalization
Daily sales aggregation (2823 records)
Prophet model training (ds, y format)
30-day future forecast generation
Trend & seasonality decomposition
Confidence interval estimation
The architecture consists of stacked LSTM layers followed by dense output layers to capture long-term dependencies and nonlinear market patterns. The model is trained on historical closing prices to predict future trends.
Project Preview