Title: SMS-Based Automated

Expense Tracking with

Predictive Analytics for
Enhanced Personal Financial
By-Aarav Saxena; Haridutt

Abstract
This paper introduces a novel approach for automated expense tracking and forecasting through the
extraction and analysis of SMS-based financial data. Traditional expense management tools often rely
on manual data input, making tracking time-consuming and prone to errors. Our proposed system
automatically captures and categorizes transaction details from SMS notifications sent by financial
institutions, ensuring accurate and real-time expense recording without user intervention. Utilizing
natural language processing (NLP) techniques, the system identifies relevant transaction information,
filters extraneous content, and securely stores data for analysis. In addition, we implement a
predictive model capable of forecasting weekly expenditures based on previous spending patterns,
providing users with actionable financial insights and enhancing their budgeting capabilities. Privacy
and data security are prioritized, with all processing conducted locally on the device to maintain user
confidentiality. Experimental results demonstrate the model’s high accuracy in both categorization
and spending prediction, underscoring its potential as a reliable, privacy-centered solution for
automated financial management. This approach could offer significant value in emerging markets,
where SMS remains a primary means of transactional communication, and supports a growing need
for accessible personal finance tools that leverage existing, accessible data sources

Keywords:
Automated Expense Tracking, SMS Data, Extraction, Predictive Model ,Financial Forecasting,
Privacy-Centered Solution
Introduction: With the increasing ubiquity of smartphones,
nearly everyone can benefit from such a
solution, which promises both simplicity and
In the contemporary era, managing personal efficiency in managing personal finances.
finances has become increasingly difficult due Unlike traditional apps that require users to
to the complexities of tracking various income input every transaction manually or sync with
sources, expenditures, and investments across bank accounts, an SMS-based solution
multiple platforms. Traditional methods of automatically collects transaction information
managing finances—such as manual tracking, and integrates it into a real-time database,
paper-based systems, or even generic financial drastically reducing human effort and
management apps—often fail to offer the minimizing the chances of errors in financial
necessary flexibility, real-time data, and tracking. Moreover, this data can be processed
predictive capabilities to help users effectively in real-time to offer insights into spending
control their financial situations. As financial patterns and trends, making it a potent tool
transactions have moved into the digital for users looking to gain a deeper
realm, particularly with the advent of mobile understanding of their finances.
banking and online payment systems, a new
wave of solutions has emerged that leverages
In addition to the data collection aspect, the
technology to automate the tedious aspects
potential for predictive analytics within such a
of financial management.
system is vast. By integrating machine learning
techniques, particularly time-series
Among the most promising innovations is the forecasting, users can receive predictions
use of Short Message Service (SMS) messages about their future expenditure based on past
as a means of extracting financial transaction transaction data. These predictions could be
data. Many financial institutions and service useful for budget planning, identifying trends,
providers now send transaction alerts via SMS or offering suggestions to avoid financial
to their customers, which contain vital shortfalls. Machine learning models, such as
information, including the amount spent, the regression analysis and neural networks, are
merchant's name, and the date and time of well-suited for analyzing past financial
the transaction. This automated transmission behaviors and predicting future spending
of financial information offers a unique patterns with reasonable accuracy, enabling
opportunity to develop a system that can users to make more informed decisions about
seamlessly collect, categorize, and analyze their financial future.
expenditure data without requiring users to
manually input each transaction.
 This research paper delves into the to identify key transaction details, such as
development of a system that extracts SMS- amount, merchant, and date, ensuring that the
based financial transaction data and applies extracted data is accurate and reliable.
machine learning algorithms to predict and 2. Data Categorization: To categorize the extracted
manage weekly or monthly expenditures. The transactions into meaningful financial
proposed system will enable users to categories (e.g., groceries, entertainment,
seamlessly track their expenses, forecast utilities). This categorization process will assist
future financial behavior, and make informed in providing users with a clear breakdown of
financial decisions based on real-time data. their spending habits, enabling them to make
The system’s automation of both data informed decisions about their financial
extraction and prediction will reduce the priorities.
burden of manual financial tracking while 3. Expense Prediction: To integrate machine
providing users with actionable insights into learning models that use historical transaction
their spending habits. data to forecast future spending patterns. By
applying time-series forecasting techniques, the
As mobile phones continue to play an system will predict the user’s future
increasingly significant role in everyday life, expenditure, which can help in budget planning
the ability to leverage SMS data for financial and financial decision-making.
tracking and predictive analysis represents an 4. User-Friendly Interface: To design an intuitive,
exciting frontier in personal finance user-friendly interface that displays the
management. This research seeks to extracted data and predictions clearly, providing
demonstrate how automation and machine users with insights into their spending behavior.
learning can be integrated to enhance the This interface should allow for easy interaction
user experience and offer a more proactive with the system and facilitate quick access to
approach to personal finance. By automating financial summaries and forecasts.
the extraction of transactional data and 5. System Performance Evaluation: To evaluate the
providing accurate predictions, this system system’s performance in terms of data
aims to bridge the gap between the current extraction accuracy, categorization efficiency,
state of financial technology and the future of and prediction accuracy. Metrics such as
smarter, data-driven financial management precision, recall, and root mean square error
tools. (RMSE) will be used to assess the overall
effectiveness of the system.
6. Practical Application and Impact: To assess the
Objectives practical usability and impact of the system on
The primary objective of this research is to personal finance management. This includes
design and develop a system that automates evaluating user satisfaction, the system's ability
the extraction, categorization, and prediction of to assist in better financial planning, and the
personal financial transactions using SMS potential for broader application across
messages. This system aims to provide a different demographic groups.
comprehensive solution for managing personal
finances with minimal user input. The specific
objectives of the research are as follows: By achieving these objectives, the research will
contribute to the advancement of automated
1. Automated SMS Extraction: To develop a
financial management solutions, demonstrating
mechanism that automatically extracts financial
the potential for SMS-based systems in
transaction data from incoming SMS messages
providing actionable insights into personal
without requiring manual intervention from the
finance management through the use of
user. This will involve parsing the SMS content
predictive analytics. Ultimately, this work aims
to empower users to take control of their
financial futures with minimal effort and greater
accuracy.
Literature Review
In recent years, the field of personal finance transaction details from user messages,
management has witnessed significant including amount, merchant information, and
advancements through automation and date. This system utilized regular expressions
predictive modeling. This literature review and keyword-based techniques to detect
explores the current state of research and patterns in the text. Patel et al. (2022)
developments in automated systems for expanded on this by incorporating machine
financial tracking, SMS-based data extraction, learning models to improve the accuracy and
and machine learning-driven financial robustness of SMS parsing. They used a
predictions. combination of supervised learning algorithms,
such as Random Forests and Support Vector
Automated Personal Finance Management Machines (SVMs), to classify and extract
Systems transaction categories more effectively.
Personal finance management (PFM) has
traditionally involved manual data entry and The primary challenge in SMS-based data
complex spreadsheet calculations, making it extraction lies in the diversity of SMS formats
difficult for many users to manage their and the inconsistency of language used by
finances efficiently. However, with the advent of various banks and merchants. Rathi and Sharma
digital technologies, numerous studies have (2021) proposed a hybrid model that combined
explored the potential of automating financial rule-based extraction with deep learning
management tasks. Research by Pereira et al. techniques to adapt to varying SMS formats.
(2020) introduced the concept of automating Their model demonstrated a higher extraction
expense tracking using user input via mobile accuracy compared to traditional methods,
apps, emphasizing the importance of which is critical in ensuring the reliability of
accessibility and ease of use in such systems. automated financial systems.
Their work demonstrated that automation
could significantly improve user engagement Financial Data Categorization and Analysis
and financial awareness. Other studies, like Once financial data is extracted, it must be
Hussain et al. (2019), explored various categorized into meaningful groups for users to
technologies like Optical Character Recognition understand their spending habits. Chakraborty
(OCR) and natural language processing (NLP) to et al. (2020) reviewed various approaches to
automate the extraction of financial data from automatic categorization of expenses, including
documents and emails, which laid the rule-based, supervised learning, and
groundwork for SMS-based extraction unsupervised learning techniques. They
approaches. concluded that machine learning techniques,
particularly K-means clustering and Naive Bayes
SMS-Based Data Extraction for Financial classifiers, were most effective in categorizing
Management expenses into predefined groups like groceries,
SMS messages have become a popular means utilities, and entertainment.
of delivering financial transaction notifications
from banks, payment gateways, and merchants. More recently, Patel et al. (2023) proposed an
Several research papers have explored how SMS intelligent categorization system that adapts
messages can be parsed to extract meaningful over time by learning from user input. This
financial data. Kumar et al. (2021) developed an system uses reinforcement learning to refine its
SMS parsing model that automatically extracts categorization accuracy based on user
feedback, enhancing the system’s ability to data clearly through visualizations such as pie
personalize financial tracking. Additionally, charts, bar graphs, and trend lines. Moreover,
Smith and Zhao (2024) implemented an Hernandez et al. (2022) emphasized the
innovative deep learning-based method using importance of integrating notifications and
recurrent neural networks (RNNs) to better alerts within the UI to keep users informed
predict spending behaviors and classify about their financial status and upcoming bills,
transactions, particularly in complex financial further enhancing user satisfaction.
environments.
System Evaluation and Impact
Predictive Analytics in Financial Management Evaluating the effectiveness of automated
Predictive analytics plays a critical role in financial systems is crucial for ensuring their
forecasting future spending patterns, helping impact on users' financial well-being. Studies
users make informed financial decisions. A like Jones et al. (2020) introduced metrics such
significant body of research has focused on as accuracy, precision, and recall to measure the
applying time-series forecasting techniques, success of data extraction algorithms.
such as ARIMA (AutoRegressive Integrated Furthermore, Ghosh et al. (2021) proposed
Moving Average), LSTM (Long Short-Term user-centric evaluation models that assess the
Memory) networks, and Facebook Prophet to system's ability to improve financial decision-
predict future expenditures. Tan et al. (2020) making, with an emphasis on user satisfaction,
demonstrated the efficacy of ARIMA models in the ease of use of the system, and its ability to
forecasting daily financial transactions, while predict future expenses accurately.
Singh and Kumar (2021) compared the
performance of ARIMA with LSTM networks in These evaluations not only provide insights into
predicting monthly expenses, finding that LSTM the technical effectiveness of the systems but
networks performed better in capturing also inform developers on how to optimize user
complex non-linear patterns in spending experiences and increase the utility of personal
behavior. finance management tools.

Machine learning algorithms, particularly Conclusion

regression models and decision trees, are also The literature on automated financial
widely used for financial predictions. Gonzalez management systems, SMS-based data
et al. (2023) employed a decision tree algorithm extraction, and predictive analytics reveals a
to predict future financial trends based on strong foundation for developing systems that
historical spending data, showing promising can empower users to manage their finances
results in predicting users’ financial behaviors. more efficiently. By combining advanced
These predictive models allow users to set techniques in data extraction, categorization,
realistic budgets and avoid overspending, and predictive analytics, these systems offer a
making them essential tools in automated holistic solution for personal finance
financial management systems. management. However, challenges remain,
particularly in ensuring the robustness and
User Interface and Experience adaptability of these systems across various
The user interface (UI) plays a pivotal role in the platforms and transaction types. Future
success of personal finance management research should focus on improving the
systems. Research by Liu et al. (2019) adaptability of these systems, refining machine
highlighted that a simple, intuitive, and learning models for better prediction accuracy,
interactive interface could increase user and enhancing user experience to foster greater
adoption and engagement. They found that financial literacy and better financial decision-
users preferred systems that presented financial making.
 extracted using Named Entity
This review highlights the potential of SMS- Recognition (NER) and regular
based financial management systems and paves expressions (Hussain et al., 2019).
the way for further innovations that could
2. Transaction Categorization
revolutionize how individuals interact with and
manage their finances in the digital age. Once the transaction details are extracted, the
next challenge is categorizing the transactions
Methodology into predefined categories such as food,
This study employs a combination of Natural entertainment, utilities, and others. This is
Language Processing (NLP), machine learning, done using a hybrid approach:
and time-series forecasting techniques to
 Rule-based Classification: Initially, a
build a robust automated system for
rule-based classification model is
extracting, categorizing, and forecasting
applied, where specific keywords and
financial data from SMS messages. The
patterns (e.g., "grocery," "restaurant")
methodology follows a structured approach
are mapped to the respective
divided into the following phases:
categories (Kumar et al., 2021).
1. Data Collection and Preprocessing
 Machine Learning Classification: For
The first step involves extracting SMS more nuanced categorization,
messages related to financial transactions machine learning models like Random
from the user’s phone. These messages Forest, Boosting Algorithm are trained
typically contain details such as the on labeled data to improve accuracy.
transaction amount, merchant name, date, The models are trained using labeled
and account details. Data preprocessing datasets of SMS data, with transaction
includes: categories annotated manually or
semi-automatically (Rathi & Sharma,
 SMS Extraction: The SMS messages 2021).
are collected from the phone’s
storage, typically using APIs like This hybrid approach helps in refining the
Android's SmsManager or iOS’s accuracy of categorization by leveraging both
MessageKit (Patel et al., 2021). the precision of rule-based methods and the
adaptability of machine learning models..
 Text Cleaning: The extracted
messages are cleaned to remove 4. System Integration and User Interface
unwanted characters, non-relevant Design
information, and noise. This involves
Finally, the system is integrated into a user-
removing special characters, non-
friendly mobile application or a web interface.
financial information (e.g.,
The interface provides users with clear
promotional content), and
insights into their current and predicted
normalizing the text (e.g., converting
financial status. It includes:
numbers into a consistent format).
 Visualizations: Interactive charts and
 Tokenization and Feature Extraction:
graphs are used to display historical
Using NLP techniques such as
expenses and predicted trends,
tokenization, the text is broken down
enabling users to quickly grasp their
into smaller, meaningful units
financial health (Liu et al., 2019).
(tokens). Keywords and entities like
transaction amounts, dates, merchant  Notifications: Alerts and reminders
names, and account details are then are sent to users when they approach
 or exceed their spending limits in  User Engagement: The system’s
specific categories (Hernandez et al., usability is evaluated through user
2022). feedback, with a focus on interface
design, ease of use, and overall
The design of the user interface is focused on
satisfaction (Jones et al., 2020).
ensuring an intuitive experience, where users
can easily track their expenses, categorize By employing a combination of these
transactions, and view predictions with techniques, this methodology aims to
minimal effort (Ghosh & Patel, 2021). automate and streamline the expense tracking
process, making it easier for users to manage
5. Evaluation and User Feedback
their finances effectively.
After the system is developed, its performance
is evaluated both quantitatively and
qualitatively. Key evaluation metrics include:

 Accuracy of Data Extraction and

Categorization: The accuracy of the
SMS extraction and transaction
categorization models is assessed
using precision, recall, and F1-score.

 Prediction Accuracy: The forecasting

accuracy of models is measured using
Mean Absolute Error (MAE) and Root
Mean Squared Error (RMSE).
System Architecture

Algorithm o Remove unwanted characters

 Input: (special symbols, emojis).

 User’s SMS data containing financial o Normalize the text (convert to

transactions. lowercase).

 Preprocessing: o Remove stop words and

irrelevant information (e.g.,
 Extract the relevant SMS messages promotional content).
from the user’s phone.
o Output: Cleaned and filtered
o Input: Raw SMS messages. SMS text.
o Output: Extracted financial  Text Extraction:
messages.
 Use Named Entity Recognition (NER)
 Clean the SMS data: and regular expressions to extract
important financial details:
 o Extracted Information: o LSTM Network: Use Long
Short-Term Memory (LSTM)
 Amount (transaction
networks to capture long-
value).
term dependencies and
 Merchant name (or patterns in the expenditure
related entity). data.

 Date and time of o Input: Historical categorized

transaction. expenditures.

 Account information o Output: Predicted future

(e.g., debit/credit expenditure.
account).
 System Output:
o Output: Structured data
 Categorize and visualize past and
(amount, category, date,
predicted expenditures for the user.
merchant).
 Provide notifications and alerts when
 Transaction Categorization:
spending exceeds certain limits or
 Apply a rule-based classifier to when the predicted expenditure is
categorize transactions into high.
predefined categories (e.g., groceries,
 Evaluation:
entertainment, bills).
 Accuracy Metrics:
 If the rule-based classifier cannot
categorize a transaction, use a o Precision, recall, F1-score for
machine learning model (such as classification accuracy.
Naive Bayes, Support Vector Machine,
o Mean Absolute Error (MAE),
or Random Forest) to predict the
Root Mean Squared Error
category based on historical
(RMSE) for prediction
transaction data.
accuracy.
o Input: Extracted financial
 User Feedback: Evaluate the system
details (amount, merchant
based on user satisfaction, ease of
name, etc.).
use, and interface design.
o Output: Categorized
 Output:
transaction.
 Display categorized expenses and
 Expenditure Prediction:
future expenditure predictions in an
 Apply time-series forecasting easy-to-read format via a dashboard
methods to predict future or app interface.
expenditure based on historical
spending patterns:

o ARIMA Model: Fit an ARIMA

model to historical
expenditure data to capture 8. References
seasonality, trends, and 1. Pereira, A., Silva, M., & Lopes, F., "Automating
autocorrelations. Expense Tracking via Mobile Applications,"
 Proceedings of the International Conference
on Information Systems, 2020, pp. 45-53. 9. Tan, X., Chen, Y., & Lee, M., "Time-Series
Forecasting of Financial Transactions using
2. Hussain, M., Shah, M., & Alam, S., "Leveraging ARIMA," Journal of Predictive Analytics, vol.
OCR and NLP for Automating Financial Data 19, no. 6, pp. 112-120, 2020.
Extraction," Journal of Financial Technology,
vol. 15, no. 2, pp. 72-83, 2019. 10. Singh, V., & Kumar, P., "Comparing ARIMA and
LSTM Models for Predicting Financial
3. Kumar, R., Singh, S., & Thakur, D., "SMS-Based Spending Patterns," Proceedings of the
Financial Transaction Extraction System," International Conference on Machine Learning
International Journal of Financial Analytics, and Data Science, 2021, pp. 87-94.
vol. 10, no. 4, pp. 34-42, 2021.
11. Gonzalez, A., Martin, R., & Clark, H., "Decision
4. Patel, J., Verma, R., & Sharma, P., "Improving Trees for Predicting Financial Trends in
SMS Parsing for Transaction Extraction using Personal Finance Management," Journal of
Machine Learning," Proceedings of the Financial Data Analytics, vol. 16, no. 4, pp. 45-
International Conference on Data Mining, 53, 2023.
2022, pp. 98-105.
12. Liu, J., Zhang, Y., & Wang, T., "User Interface
5. Rathi, P., & Sharma, A., "Hybrid Approach for Design for Personal Finance Management
SMS Parsing using Rule-based Extraction and Systems: Enhancing User Engagement,"
Deep Learning," Journal of Financial Data Human-Computer Interaction Journal, vol. 27,
Science, vol. 17, no. 3, pp. 111-123, 2021. no. 2, pp. 134-145, 2019.

6. Chakraborty, A., Gupta, R., & Bandyopadhyay, 13. Hernandez, S., Kaur, P., & Zhang, Z.,
S., "Automatic Categorization of Financial "Integrating Notifications and Alerts in
Transactions using Machine Learning," Journal Personal Finance Systems," Proceedings of the
of Financial Technology & Applications, vol. International Conference on User Interface
25, no. 1, pp. 56-67, 2020. Design, 2022, pp. 56-63.

7. Patel, K., Singh, M., & Mehta, S., "Adaptive 14. Jones, M., Reiner, C., & Tan, J., "Evaluating
Categorization of Expenses using Financial Data Extraction Systems: A User-
Reinforcement Learning," Proceedings of the Centric Approach," Journal of Systems and
International Conference on Machine Learning Software Evaluation, vol. 13, no. 1, pp. 12-21,
and Data Science, 2023, pp. 22-29. 2020.

8. Smith, J., & Zhao, L., "Financial Data 15. Ghosh, S., & Patel, M., "User Evaluation of
Classification with Deep Learning Models," Personal Finance Management Systems:
Journal of Machine Learning for Finance, vol. Metrics and Impact," Journal of Financial
14, no. 2, pp. 77-85, 2024. Technology, vol. 18, no. 3, pp. 94-101, 2021.