Current research often focusses on the use of Digital Twins for data collection and visualization and Machine Learning for data analysis to develop prediction models. However, these research lack discussion on how data to develop predictive models/twin needs to be selected and how they contribute to the models’ accuracy and effectiveness. In this paper the authors focus their attention on how the data needs to be selected for the development of accurate and cost-effective prediction models. The paper developed two machine learning models, one containing redundant data and another with redundant data combined as single data points. During testing, both models achieved similar accuracy of 0.86, highlighting that redundant data did not add to the accuracy of the predictive model/twin. The results also show that collection of redundant variables can be eliminated to reduce the cost of data capture and storage.