Feature: Improve forecasting models and model selection process

[chripiermarini]

Description

The current system relies on simple baseline forecasting models (naive, seasonal, rolling average), and model selection is based on a single metric (WAPE).

The goal of this issue is to improve both:

• the forecasting model library
• the robustness and flexibility of the model selection process

---

Goals

• Expand the set of forecasting models
• Improve evaluation and selection logic
• Maintain modularity and extensibility of the forecasting layer

---

Proposed Improvements

Add more advanced models, such as:

• Exponential Smoothing (ETS)
• ARIMA / SARIMA
• (Optional) tree-based models (e.g. gradient boosting)

Extend evaluation metrics:

• RMSE
• MAE
• MAPE
• WAPE (already implemented)

Improve model selection:

• Allow configurable metric selection
• Explore multi-metric ranking or weighted scoring
• (Optional) introduce rolling / time-based validation

---

Acceptance Criteria

• At least 1–2 new forecasting models implemented
• Model selection supports configurable metric input
• Evaluation is consistent across all models
• Code remains modular and easy to extend