A retailer built a simple linear model to predict sales from price, and the residual plot indicates a clear non-linear pattern. Which actions should the ML engineer take next to improve prediction accuracy? (Choose three.)

A clear non-linear pattern in the residual plot indicates that the linear model is not capturing the underlying relationship in the data. 1. Replace the linear model with a non-linear model: This is a direct approach to address non-linearity. Models like decision trees, random forests, or neural networks can inherently capture complex, non-linear relationships. 2. Inspect the input features and apply appropriate non-linear transformations where needed: If the non-linearity is due to a specific feature, transforming that feature (e.g., using log, square, or inverse transformations) can make the relationship more linear, allowing the original linear model (or a new one) to perform better. 3. Split the data and build two separate models for different regions of the data: If the non-linear pattern suggests different linear relationships in different data ranges (e.g., one slope for low prices and another for high prices), building separate models for these distinct regions can effectively capture the overall non-linearity. Downsampling or offsetting input values are generally not effective strategies for addressing non-linearity in residual plots; they might discard valuable information or arbitrarily shift data without addressing the underlying functional form.