In the first part of a series on fraud detection using Neo4j and graph data science, a sample graph dataset from a real-world peer-to-peer platform is introduced, featuring anonymized user accounts, transactions, and identifiers such as credit cards and devices. The dataset is structured to explore potential fraud patterns, with a small percentage of accounts flagged for fraud based on chargeback events and manual review. Initial analysis reveals that fraud accounts are not well-connected, suggesting possible limitations in the current labeling approach. To identify fragmented identities of fraudsters, community detection methods like Louvain are employed to partition the graph into well-connected groups. This exploratory analysis uncovers suspicious patterns, such as flagged users transferring money to non-flagged users with shared identifiers, which may indicate fraud risks. The article concludes with a preview of using these patterns to refine fraud detection in future installments of the series.