I am using Amazon Forcast service to predict sales in 1 year and have difficulty validating the data. So I have a scenario like this:
- Datasets : TARGET_TIME_SERIES data from January to December 2019 with attributes: item_id, location, timestamp, demand | ITEM_METADATA data with attribute item_id and type_id.
- Predictor : Forecast horizon 10 | Forecast frequency 1 week | Algorithm : AutoML | Optimization metrics : rmse, wape, mase, mape, averageWQL | Forecast dimensions : location | Number of backtest windows : 1 | Backtest window offset : 10 | Forecast type : 0.50, 0.60, 0.70, 0.90, mean.
I have a problem when the forecast results use a variety of different metrics but none of the chart patterns match the data when compared to the actual data for 2020. Here I attach the graph.
Can you help me solve this problem? Please give me instructions and suggestions, because I have tried various ways but when the validation results are still not appropriate. thank you
