Predict if a transaction is fraudulent or not. The file "transactions.csv$"$ contains all transactions. The file solely contains numerical input variables resulting from a PCA transformation. Original features cannot be supplied due to privacy concerns. The main components derived with PCA are features V$1,$ V$2,...,$ V$28$; the only features that have not been changed with PCA are 'Time' and 'Amount'. The 'Time' feature contains the number of seconds that have passed between each transaction and the first transaction in the dataset. The feature 'Amount' represents the transaction Amount; this feature can be utilised for cost$-$sensitive learning, for example. The answer variable is called 'Class,$\text{'}$ and it has a value of $1$ if there is fraud and $0$ otherwise. Examine the dataset. Do you believe the dataset is balanced? Evaluate and compare the XGBoost and RandomForest algorithms without and after data balancing with SMOTE. For hyperparameter optimisation, compare performance using the F$1$ score and apply $5-$fold cross$-$validation. Discuss which of the four situations $($Random Forest, XGBoost, Random Forest $+$ SMOTE, XGBoost $+$ SMOTE$)$ performs the best?