A consumer goods company has full sales history for many product variants and currently uses ARIMA models. They want to forecast demand for a new product that will be launched soon. Which approach should a Machine Learning Specialist use?

DeepAR is a neural network-based forecasting algorithm that is particularly effective for forecasting demand for new products or products with limited historical data. It learns patterns across a large collection of related time series, allowing it to generalize and make predictions for new items by leveraging information from similar existing products. Training a custom ARIMA model specifically for the new product is not feasible because ARIMA models require historical data for the specific series they are forecasting, which a new product lacks. K-means clustering is an unsupervised learning algorithm used for grouping data points, not for time series forecasting. XGBoost is a powerful supervised learning algorithm but is typically used for tabular data prediction and would struggle with the time-series nature and lack of historical data for a new product without significant feature engineering.