Film breakage is a critical defect in plastic film manufacturing, directly impacting product quality and customer satisfaction. In smart manufacturing environments, defect prediction is complicated by large datasets and diverse production factors, a challenge that remains underexplored in the literature for this specific industry. This study addresses this gap by presenting a comparative analysis of machine learning approaches for feature selection and breakage prediction. First, this research introduces two embedded feature selection methods: Decision Tree (DT) and Random Forest (RF), leveraging the Mean Decrease Impurity (MDI) metric. Second, it compares the predictive performance of three regression models - Decision Tree Regression, Random Forest Regression, and Artificial Neural Networks (ANN) - using the selected features. The experimental results demonstrate that the RF model achieves the highest predictive accuracy while feature selection significantly reduces computation time with 5 prominent features selected by DT. These findings highlight the potential of integrating feature selection with machine learning to improve defect prediction, optimize production processes, and ensure higher quality in plastic film manufacturing.