What are some common metrics used to evaluate the performance of a machine learning model, and how do you determine which metric to use based on the problem?

During my final year project, I developed a machine learning model to classify emails as spam or not spam. I initially used accuracy as my metric, but then I realized that my dataset was imbalanced, with far more legitimate emails than spam. This led to a misleadingly high accuracy score. To address this, I switched to using precision and recall, which helped me understand the model's effectiveness in correctly identifying spam without misclassifying too many legitimate emails. This experience taught me the importance of choosing the right metrics based on the nature of the data.

While volunteering for a local nonprofit, I helped analyze survey data to assess community needs. We created a predictive model to identify which services would be most beneficial. Initially, I focused on accuracy, but I soon realized that it didn’t capture the model's effectiveness in prioritizing the most vulnerable populations. I learned to incorporate metrics like recall to ensure that our model accurately identified individuals in need. This experience highlighted the necessity of adjusting metrics based on the social impact of our predictions.

In my internship at a tech startup, I worked on predicting customer churn using regression analysis. I used metrics like Mean Absolute Error (MAE) and R-squared to evaluate model performance. I found that MAE provided a clearer picture of average prediction errors, which was crucial for understanding customer behavior. This experience helped me appreciate how different metrics can change the narrative of model performance and how they should be selected based on the specific business goals we aimed to achieve.