On blockchain platforms, individuals can engage in transactions and authentication processes without undergoing identity verification. This characteristic presents challenges for third-party service providers, such as Decentralized Applications, in accurately identifying and targeting their customer base. Consequently, the demand for detecting potential users has surged recently. However, most solutions today rely on off-chain data such as those on public forums and social media. Although these methods are straightforward, they demand significant time and effort and exhibit limited scalability and flexibility. In this research, we propose a novel methodology to identify potential users for a given lending decentralized application. The proposed approach involves developing a machine learning-based solution that leverages on-chain transaction data, rather than solely depending on the rule-base. Specifically, we employ blockchain mechanisms to gather and analyze user data from DApps. Based on the identified patterns, we formulate hypotheses and apply machine learning techniques to train the model. The focus of this research is on predicting potential users for DApps operating on EVM (Ethereum Virtual Machine) blockchain networks, with evaluations conducted on the seven largest chains: Ethereum, BNB Chain, Polygon, Fantom, Avalanche, Optimism, and Arbitrum. The proposed detection method demonstrates promising results, achieving an accuracy rate and F-2 score of approximately 89% and 72% respectively. Furthermore, the customer acquisition cost (CAC) is estimated to be around $3,133, which is superior to the strategies employed by DApps like 1inch and three times more cost-effective than Uniswap.