Time-Series Forecasting Model Development

I have a set of time-stamped observations and I need a robust machine-learning solution that can forecast future values with solid accuracy. The end goal is an automated prediction/forecasting model that I can retrain as new data arrives and easily integrate into a larger Python pipeline.

Here’s what I need from you:

• Exploratory analysis of the raw time-series data, spotting seasonality, trends, and anomalies.
• Thoughtful feature engineering (lags, rolling stats, calendar effects, exogenous variables if useful).
• Model selection and training – I’m open to traditional approaches such as ARIMA/Prophet as well as more advanced architectures like Gradient Boosting or LSTM; choose what performs best and explain why.
• Rigorous evaluation on a hold-out set with clear metrics (MAPE, RMSE, or similar) so I can understand real-world performance.
• Clean, well-commented Python code (Jupyter notebook or .py scripts) plus a brief README describing setup, retraining, and inference steps.
• Optional but appreciated: a lightweight way to serve the model (e.g., FastAPI endpoint or batch script) so it can slot straight into production.

I’ll provide the dataset and any domain context you need right after kickoff. If you have experience with pandas, NumPy, scikit-learn, statsmodels, TensorFlow/PyTorch, or Prophet, you’ll be right at home. Accuracy, clarity, and reproducibility are more important to me than flashy visuals, but a concise plot or dashboard that helps explain the results would be a bonus.

Let me know what modeling approach you’d start with, how long you’ll need to deliver the first working prototype, and any assumptions you’d like me to confirm before we begin.