We will predict how many meals are served each day for every school, separately for Breakfast and Lunch. This helps us see year-over-year changes in participation and plan production to reduce waste and control costs. The same data will also support “cost per meal” analysis.
Data we'll use: fact_production.csv
Which has one row per school * Date * Meal with fieds like served_total, offered_total, planned_total, production_cost_total and school info.
Model:
- Start with simple baselines: use yesterday value and use last last week same weekday
- Try classic time series models(SARIMA/Prophet)
- We will use advanced LSTM model to compare with simple model
- Coming to Features , simple calendar info(day of week, month), past values (7/14/28 days), rolling averages, and school attributes.
- We are not using POS/Menu data as of now , if we want we will add those later.
We will check accuracy using rolling time-based split, evaluate MAPE and MAE and show charts of actual vs forecast.
We will predict how many meals are served each day for every school, separately for Breakfast and Lunch. This helps us see year-over-year changes in participation and plan production to reduce waste and control costs. The same data will also support “cost per meal” analysis.
Data we'll use: fact_production.csv
Which has one row per school * Date * Meal with fieds like served_total, offered_total, planned_total, production_cost_total and school info.
Model:
We will check accuracy using rolling time-based split, evaluate MAPE and MAE and show charts of actual vs forecast.