As　the　financial　industry　continues　to　open　up,　numerous　multi-party　data　modeling　cases　have　emerged.　Given　the　high　value　and　confidentiality　of　financial　data,　the　modeling　process　faces　significant　security　risks,　including　the　potential　leakage　of　critical　financial　information,　such　as　user　privacy.　Ensuring　that　data　remains　＇usable　but　not　visible＇　during　the　modeling　process　has　become　a　critical　technical　challenge　in　this　field.　Privacy　computing　seeks　to　enable　data　analysis　and　computation　without　external　exposure,　transforming　and　unlocking　data　value　while　fully　safeguarding　both　data　and　privacy.　Homomorphic　encryption　is　a　key　privacy　computing　technology　that　encrypts　raw　data　and　allows　specific　operations　to　be　performed　on　the　ciphertext.　The　decrypted　results　of　these　operations　are　equivalent　to　those　performed　on　plaintext　data.　Compared　to　other　privacy　protection　methods,　such　as　differential　privacy,　federated　learning,　and　k-anonymity,　homomorphic　encryption　stands　out　as　a　provably　secure　cryptographic　solution　for　cloud-based　privacy　computing.　It　enables　fraud　detection　models　in　cloud　computing　to　perform　encrypted　polynomial　calculations　on　ciphertext　data　without　accessing　users＇　real　data,　making　ciphertext-based　fraud　detection　a　prominent　research　area.　To　address　the　issue　of　privacy　data　leakage　in　multi-party　data　model　construction,　this　paper　proposes　a　novel　secure　multi-party　computation　model　based　on　homomorphic　encryption.　We　introduce　a　homomorphic　encryption-based　machine　learning　algorithm,　evaluating　its　accuracy　and　efficiency　on　real　datasets,　demonstrating　its　effectiveness　and　potential　for　practical　applications.　©　2024　IEEE.