Sales prediction time-series model

An applied time-series learning project using a sales dataset with date, store, product, and number-sold fields.

Workflow

• Selected one store and product series from the larger dataset.
• Scaled the target values and used time-aware splits rather than randomly shuffling the sequence.
• Trained an LSTM model and compared its prediction error with a simple threshold.

The public notebook reports an unscaled mean absolute error of 0.13 for the sampled series. The old portfolio rounded this to 0.1.