On-Device Financial Data Management Using ML and NLP: A Privacy-Focused Approach

Personal finance management has become more difficult due to the exponential rise of digital financial services, with SMS-based notifications being a crucial channel for transaction updates. However, the unstructured nature of SMS data poses challenges for users in organizing financial information effectively. This paper introduces a novel, privacy-preserving on-device financial data management system that makes use of Natural Language Processing (NLP) and Machine Learning (ML). The proposed solution incorporates three core functionalities: SMS transaction classification, data extraction, and data categorization, complemented by a chatbot interface for user interaction. A Random Forest classifier achieves a remarkable 99.15% accuracy in transaction classification, while a Bi-LSTM-CRF model, combined with BERT embeddings, attains 99.61 % accuracy in extracting transaction details. Transactions are categorized into spending categories facilitating detailed expense analysis. Optimized for mobile deployment, the system ensures data privacy and efficiency. This work bridges a critical gap in personal finance management by enabling users to securely track, analyse, and gain insights into their financial behaviour, laying the groundwork for future enhancements like real-time alerts and predictive analytics.