A retail ML specialist is forecasting daily sales for a store using 10 years of historical daily sales. About 5% of days have missing values. The forecast errors are highest around seasonal events, where predictions are consistently biased. Which two actions should the specialist take to try to improve model performance? (Choose two.)

Adding features that describe sales periods (holidays, promotions, special events) directly addresses the problem of biased predictions around seasonal events. These features provide the model with crucial information about factors influencing sales spikes or dips, allowing it to learn and account for these patterns. Applying smoothing or modeling techniques to correct for seasonal variation is also effective because it helps the model better capture and predict recurring patterns in the data, thereby reducing forecast errors during these periods. Aggregating sales across nearby stores might increase data volume but could introduce noise if the stores have different sales patterns. Changing forecast frequency to weekly reduces granularity and might obscure important daily patterns. Filling missing values with linear interpolation is a common imputation technique but doesn't address the core issue of biased predictions during seasonal events.