How would you approach selecting the right evaluation metric for a regression problem?

In a college project, I developed a regression model to predict housing prices based on various features like location, size, and amenities. I initially considered multiple evaluation metrics but decided to focus on Mean Absolute Error (MAE) because it provided a clear interpretation of average prediction error. This was particularly helpful, as stakeholders wanted to understand how much off my predictions could be in real terms. By analyzing MAE, I was able to fine-tune my model effectively, which taught me the importance of aligning metrics with project goals.

While volunteering for a non-profit organization, I was tasked with predicting the amount of funds raised during events. I used a regression model and opted for R-squared as my evaluation metric. This was because the organization wanted to grasp the overall effectiveness of different fundraising strategies. By presenting the R-squared value, I could demonstrate how well our model explained the variation in fundraising outcomes. It helped the organization make data-driven decisions about future events and optimize their efforts, showcasing how the right metric can influence organizational strategy.

In my first job, I worked on a sales forecasting regression model. I initially used Mean Squared Error (MSE) to evaluate model performance, as I wanted to penalize larger errors more heavily. However, I realized that this approach wasn't always suitable for our business context, where smaller, consistent errors were preferable over sporadic large ones. After discussing with my team, we switched to MAE, which provided a more balanced view of our predictions. This experience taught me the importance of engaging with stakeholders to determine the most appropriate metric based on business needs.