Capstone Projects

Regression Model Testing, Comparison, and Selection for Predicting Manufacturing Labor Hours at SPX Transformers Solutions

Program: Data Science Master's
Location: Wisconsin (onsite)
Student: Benjamin Wildmon

This study created a multiple linear regression model to predict the direct labor hours required to produce engineered-to-order power transformers. The predicting labor values are used in both evaluating the cost to manufacture a particular transformer design as well as to ensure the production schedule can adequately accommodate the necessary labor. Hundreds of variables, representing a multitude of attributes and characteristics describing the design of a potential transformer, were evaluated using stepwise regression variable selection methods. Variance inflation factors were calculated to test for and understand any correlation or covariance issues. Mean squared error was used as the primary measurement of prediction error and was compared to the error values to a previously existing model to ensure the prediction accuracy levels were maintained or improved.