Hybrid AI-Based Intelligent Personal Finance Advisor for Behaviour-Aware Budgeting

Personal financial mismanagement affects a large proportion of individuals globally, with many lacking tools for intelligent expense tracking, budget adherence, and goal-oriented savings. Existing applications often rely on manual categorization and rule-only logic, leading to poor scalability and limited personalization. This paper presents an AI-powered personal finance advisor that combines machine learning–based expense categorization with a FastAPI backend and a Flutter mobile frontend. The proposed system uses a hybrid approach: rule-based merchant matching for known vendors and a Multinomial Naive Bayes classifier with TF-IDF text vectorization for unseen transactions across twelve expense categories. Budget recommendations follow the 50/30/20 rule; a rule-based analytics engine delivers spending insights, anomaly detection, and savings tips; and an intent-based chatbot answers natural language queries on spending, income, savings, and goals. Authentication is enforced via JWT with bcrypt password hashing; user corrections are stored for model retraining. Experiments on a transaction dataset with 16 seed samples plus user feedback show categorization accuracy of 91.8%, precision 89%, recall 90%, and F1 score 89.5%. The REST API exhibits average response time of 180 ms and user satisfaction of 87% in pilot testing. The main contribution is a lightweight, explainable, and privacy-conscious mobile finance system integrating ML categorization, rule-based insights, and conversational support in a unified architecture.