Neural Network Model with Interactive Prediction Profiler in Action.
Situation: A company produces white polymer outdoor furniture with a process yield at around 96%. They are using Least Squares Regression to model their input and output variables. Not all statistical assumptions have been thought through and many of them are not met causing inaccurate predictions in yield.
Task: The company is not confident in their results. You suspect this may be due to over-fitting and leads to sub-optimal decisions. Along with that problem, there has been an increase in the cost of the major input factor. Your task is to develop a more robust model with an interactive prediction profiler to estimate the variable change effects.
Action: You use Neural networks to predict the response variable using a flexible function of the input variables. Neural networks with validation can be good models to use when you do not necessarily need to describe the functional form or the exact relationship between the inputs and the response.
Result: Your new Neural network model has a greater R2 than the regression model, meaning that more variation in the response variable can be explained by the predictors. The associated interactive prediction profiler allows you to answer what-if questions and identify the optimal and minimum settings.
Next Steps: This satisfies the company’s initial concerns and they are delighted with the interactivity of the prediction profiler. Congratulations, you are promoted to Business Improvement Manager and assigned a team to tackle many other important projects.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.