An Investigation into the Impact of Training Data Variations on Accuracy and reliability in Machine Learning Models