If 60-40% partitioning rates are used for training and validation, how many total observations exist in the full data set? How would that number change if it was 70-30% partitioning, 80-20% partitioning?

Bank of Fargo uses 2 predictive algorithms to evaluate credit card transactions for predicting fraudulent activity. It costs company much higher if they miss any fraudulent transaction than if any transaction is accidently identified as fraudulent, even it was not (100 times higher). Cost matrix for the binary classification outcomes is provided below (F: Fraud, NF: Non-Fraud- NOTE: Fraud cases are considered as Positive in this application)