ExpendiPredictor: Personal Expense Forecasting System with Behavioral Spending Analysis

Abstract

"In recent decades, people have frequently used digital money rather than physical money. According to the World Bank, 76% of the population owns a bank account. People can track the transactions that go through their bank account. But it had to be done manually. Most people do not have time to regularly check their bank accounts. This will result in higher expenses than savings.

Due to the data limitations in the chosen domain, SARIMAx model which is a statistical method was chosen because statistical methods don’t require vast amount of data to train a model. Since every person’s spending behaviour is different from each others, can not use a common system that pre-trained with a huge dataset. And using only single feature won’t be sufficient to give an accurate prediction since there are lot of factors affecting the spending pattern. To address that, two extra features were chosen to make the final predictions and those extra features are generated by using two SARIMAx models since in real life, future values won’t be available. Altogether with three SARIMAx models, the predictions are made and the system can successfully identify the spending pattern of an individual.

The evaluation of prediction results cannot solely rely on standard evaluation metrics due to the incorporation of two additional predictive features. Alongside traditional metrics, the Mean Absolute Error (MAE) of the prediction results is determined to be 397. These additional features pertain to the presence of expenses and extreme expenses. Notably, the system successfully predicted 3 out of 6 extreme expenses and 8 out of 10 instances with no expenses. Considering the absence of prior research on forecasting personal expenses using individuals' bank account data, these test outcomes underscore the promising potential of the proposed system."