Demand forecasting is the starting point of supply chain planning. SIMCEL uses historical demand (Base Demand) as input and offers multiple forecasting methods to project future demand.
Available forecasting models
| Model | Type | Description |
|---|
| Prophet | Time series | Meta’s algorithm that decomposes data into trend, seasonality, and holiday components |
| STL | Time series | Seasonal-Trend decomposition using Loess, separating trend, seasonal, and residual components |
| LightGBM | Machine learning | Gradient-boosting framework with leaf-wise tree growth. Faster training than XGBoost with comparable or improved accuracy |
| XGBoost (XGB) | Machine learning | Extreme Gradient Boosting for large/complex datasets, building ensembles of weak decision trees |
| AdaBoost (ADA) | Machine learning | Adaptive Boosting combining multiple weak learners by iteratively adjusting weights |
| Random Forest (RF) | Machine learning | Ensemble of multiple decision trees on randomly selected subsets of features and data |
| Extra Trees (ETR) | Machine learning | Similar to Random Forest but with different tree construction methodology |
| Decision Trees (DT) | Machine learning | Tree-like models where leaf nodes represent demand forecasts |
| Naive | Baseline | Uses previous period values (lag-1, last observation, or seasonal lag) as the forecast. Often used as a comparison baseline |
Forecast strategies
Recursive vs. non-recursive forecasting
| Strategy | How it works | Trade-offs |
|---|
| Recursive | Model predicts the next value; predictions are used as new training data for subsequent steps | Faster but bias compounds over time, lowering accuracy |
| Non-recursive | N different models are trained for N steps ahead | More time-consuming but provides better results |
Forecast settings
Access forecast settings by clicking + Plan and then Open Forecast Settings.
Forecast aggregation
Controls the granularity at which forecasts are generated. This involves a strategic balance:
| Level | Pros | Cons |
|---|
| Higher aggregation | Faster forecasting, denser data representation | Reduces precision, loses nuances and patterns |
| Finer granularity | More detail, higher precision | Requires more computational resources, may create discontinuities with insufficient data points |
Strike a balance that preserves critical information while ensuring computational feasibility. Excessive aggregation obscures vital trends; overly detailed data creates unreliable forecasts.
Explore products/customers
A feature within forecast settings that lets you review actual demand for selected products/customers from the most recent year. Useful for:
- Identifying customers or products removed from the active list but not updated in master records
- Determining whether items should be delisted
.svg?fit=max&auto=format&n=NwcmN3Uz2rKt7z_Z&q=85&s=5062a9f3b73c4eb9710e4e28d530d3dd)
