Abstract:
The increasing complexity of personal finance, combined with limited financial literacy, makes
it difficult for individuals to manage their budgets, spending, and investments effectively.
Existing tools often provide static templates that lack personalization or real-time intelligence.
In countries like Sri Lanka, many rely on outdated apps or informal methods that overlook
dynamic behaviors such as SMS-based spending. This gap hinders efficient financial decision
making, especially for students and middle-income earners.
SpendSense addresses this challenge through an AI-powered personal financial advisor that
combines machine learning and NLP to deliver personalized budget plans, anomaly detection,
and stock insights. Built using a Flask backend and React frontend, it leverages a multi-output
XGBoost model for expense prediction, Isolation Forest for detecting anomalies in SMS
transactions, and Prophet for stock forecasting. Structured SMS data is extracted using Gemini
1.5 Flash. The system includes an interactive dashboard with real-time visualizations, chatbot
based advice, and secure user sessions.
Evaluation results show high accuracy, with the expense predictor achieving R² scores above
0.90 and MAE as low as 21.20. Stock forecasts had MAE between 48.86 (GOOGL) and 140.86
(TSLA). The anomaly detection model flagged 4 out of 50 messages (8%) effectively. These
results validate SpendSense as a reliable, explainable, and adaptive solution for personal
finance management.