Pellet quality is a major concern for commercial feed mills as it influences the production efficiency of pelleted feed, as well as downstream animal performance. The Pellet Durability Index (PDI) is commonly used to measure pellet quality. However, since numerous variables can affect PDI, it is challenging for commercial feed mills to consistently achieve the desired PDI. In this study, machine learning was applied to predict PDI, using a dataset containing 55 features collected from a commercial feed mill. Feature engineering approaches, including feature creation and feature selection were applied before modeling to explore the potential patterns in the features. Based on the engineered features, the Support Vector Regression model is emerging as the best-performing model. Several features consistently emerged as key predictors across multiple models, including Dehydrated Bakery Meal (%), Ambient Temperature (°C), Crude Protein Content (%) × Starch Content (%), and Conditioning Temperature (°C) × ADF (%). This study describes a pipeline for predicting pellet quality using a large set of features and provides insights into how various factors impact high-quality pellet production.

Read the full article in the Journal of Animal Science.