Traditionally, rule-based systems have been the primary instrument for detecting fraud in today's financial systems, with fraud specialists defining the rules based on prior instances and outcomes. If a new transaction meets one or more of the previously established criteria, an alert is triggered, indicating that the new transaction may be fraudulent. For previously known fraud tendencies, the rule-based method is effective. Before adding a new rule to the current rule set, a sufficient number of fraudulent transactions must have happened that fit the rule. During this time span, fraud techniques may evolve, resulting in the induced rule expiring. Thus, the emphasis should be on using prior transactions that follow a rule-based approach in conjunction with an unsupervised method that detects previously unknown fraud activity. There is a need to use fraud detection systems that are capable of keeping up with the cardholder's updated spending behaviour. The detection process's goal is to identify as much fraud as possible while reducing the false positive rate, which has a negative effect on cardholder satisfaction as the cost of providing more false alarms increases. To accomplish this approach, the threshold value is determined at the account level of the cardholder by evaluating the probability sequence of previous and new incoming transactions. Additionally, identified fraudulent transactions are labelled in the database for future analysis in the event that additional assessment is required.