‍Unlocking Secure Forecasting in Financial Services with VENumML

‍

Time series forecasting is a cornerstone of operational efficiency in industries like finance, healthcare, and retail. Whether predicting transaction volumes at ATMs or estimating energy demand, accurate forecasts can drive better decision-making. However, for industries handling sensitive data, such as financial services, the challenge lies in leveraging this data securely without compromising privacy. Vaultree’s VENumML (Vaultree Encrypted Numbers Machine Learning), featuring Phineus, our time series module, addresses this challenge with a groundbreaking approach to encrypted forecasting.

‍

Real-World Challenge: Forecasting ATM Transaction Volumes

‍

Imagine a private ATM service provider managing millions of transactions daily, ranging from deposits and withdrawals to balance inquiries. To ensure every ATM in its network is adequately stocked with cash, the service must forecast daily transaction volumes. However, this data is highly sensitive. Beyond containing private customer details, transaction patterns are considered proprietary business information. Regulatory frameworks such as GDPR in the EU or the Gramm-Leach-Bliley Act in the U.S. require stringent measures to safeguard this information, making it challenging to perform analysis without risking breaches.

‍

The Solution: Using Vaultree’s VENumML library, encrypted forecasting can now become a reality. The Phineus demo showcases how time series forecasting models can operate directly on encrypted data, preserving privacy while delivering actionable insights.

‍

Breaking Down the Phineus Demo

‍

The Phineus demo demonstrates encrypted forecasting for ATM transaction volumes using advanced time series modelling techniques. To do so, we used synthetic time series data mimicking realistically transaction volumes for multiple banks. We made sure to include:

• Trends: Long-term growth in transaction volumes over time.
• Seasonal Patterns: Regular fluctuations due to holidays, pay cycles, and other periodic events.
• Random Noise: Day-to-day variability in transaction volumes.
• Spikes: Sudden, unexpected surges from events like large deposits or withdrawals.
• Random Walks: Unpredictable changes reflecting external factors like economic shifts.

The generated time series data is encrypted using VENumpy, Vaultree’s proprietary homomorphic encryption library. This ensures that all information remains protected throughout the entire forecasting process. The encrypted data is processed using Fourier transforms to decompose the time series into its underlying components, followed by a linear regression model to predict future transaction volumes – all without decrypting the data! Privacy is never compromised.

‍

At Vaultree, privacy is not enough. We want to make sure developers and enterprises using our technology won’t need to compromise on accuracy and performance. This is why we showcased in our demo how our encrypted model performs against industry-standard tools like Prophet. And we are so excited about the results!

‍

‍

Key Benefits of Phineus and VENumML

‍

The Phineus demo highlights the transformative potential of VENumML for secure forecasting:

• End-to-End Encryption: Data remains encrypted during all stages—preprocessing, training, and inference.
• Compatibility with Advanced Models: Supports Fourier transforms, regression, and more, ensuring state-of-the-art forecasting capabilities.
• Real-World Scalability: Optimised for large-scale, complex datasets typical of financial and operational use cases.
• Regulatory Compliance: Meets stringent privacy standards, enabling seamless integration into sensitive workflows.

Secure time series forecasting is more than a technical achievement—it’s a business enabler. By removing barriers to analysing sensitive data, Phineus empowers industries like finance to:

• Optimise cash management at ATMs, reducing downtime and operational costs.
• Safeguard proprietary transaction data from breaches or misuse.
• Build trust with customers and regulators by ensuring robust data privacy.

‍

Get Started with Phineus and VENumML

‍

Ready to explore the future of encrypted forecasting? The Phineus demo notebook is now available as part of the VENumML open-source library. It walks you through every step, from encrypting time series data to generating secure forecasts. Designed for developers and data scientists, this demo is your gateway to unlocking privacy-preserving innovation.

‍

Check out our GitHub repository to dive in. We’re excited to see how you’ll use VENumML to revolutionise your industry! If there is anything you need, let us know via Github issues or our Support portal.

‍