Behavioral Biometrics for Smartphone Authentication Using Hybrid ML and TOPSIS

William Wook^*
*Correspondence: William Wook, Department of Biostatistics, University of Helsinki, Helsinki, Finland, Email:

Introduction

In an increasingly mobile-centric world, ensuring secure access to smartphones has become critical. Traditional authentication methods such as PINs, passwords, and fingerprints face challenges such as being easily guessable, spoofable, or vulnerable to theft. Behavioral biometrics, which authenticates users based on their unique usage patterns like typing rhythm, swiping behavior, and motion dynamics, offers a promising solution. This technique not only provides continuous and passive verification but also enhances user convenience and security. The integration of hybrid machine learning (ML) algorithms with decision-making models such as TOPSIS (Technique for Order of Preference by Similarity to Ideal Solution) further strengthens the accuracy and reliability of these systems. This paper explores a robust framework combining behavioral biometrics, machine learning, and TOPSIS for secure and intelligent smartphone authentication. Behavioral biometrics leverages the unique, repeatable patterns of human-device interaction to verify identity. These may include touch pressure, swipe speed, device orientation, typing cadence, and motion sensors data (gyroscope and accelerometer readings). Unlike static biometrics such as facial recognition or fingerprints, behavioral traits are difficult to replicate and enable continuous authentication [1].

Description

In this framework, raw behavioral data are first collected and preprocessed from smartphone sensors and user interactions. Hybrid machine learning techniques such as ensemble models that combine decision trees, Support Vector Machines (SVM), and neural networks are then used to detect and classify user patterns. To enhance decision-making and reduce false positives/negatives, the TOPSIS method is integrated into the classification pipeline. TOPSIS ranks authentication confidence levels based on proximity to an ideal user profile while considering multiple weighted criteria. This approach allows the system to handle real-world variability in user behavior more effectively. For instance, if a user is walking or under stress, the model can still determine their authenticity by comparing current input to both ideal and non-ideal patterns. The hybrid use of ML and TOPSIS boosts the modelâ??s performance in terms of accuracy, speed, and robustness against spoofing. Security evaluations using benchmark datasets or real-user trials show that the framework offers high precision, low Equal Error Rates (EER), and real-time response. Importantly, it preserves user privacy, as behavioral data are difficult to reverse-engineer into personal identifiers. This makes the model particularly suitable for mobile banking, e-commerce, secure messaging, and enterprise-grade applications [2].

Behavioral biometrics offers a cutting-edge approach to user authentication by analyzing the distinct behavioral patterns individuals exhibit while interacting with their smartphones. Unlike traditional biometrics that rely on physical traits (such as fingerprints or facial recognition), behavioral biometrics examine how a user performs actions such as how they type, swipe, hold the device, walk, or even interact with on-screen elements. These subtle patterns are nearly impossible to mimic, making them highly secure and ideal for continuous, passive authentication. The proposed framework collects raw data from multiple smartphone sensors, including accelerometers, gyroscopes, touchscreens, and GPS, to capture diverse behavioral traits in real-time. This data is then cleaned and normalized to remove noise, inconsistencies, and device-specific variations. After preprocessing, the cleaned dataset is passed through a hybrid machine learning pipeline that employs a combination of algorithms such as Support Vector Machines (SVM), Random Forests, K-Nearest Neighbors (KNN), and Neural Networks. The hybrid model is particularly advantageous as it leverages the strengths of multiple classifiers, resulting in more robust and accurate predictions than any single model [3].

To further improve decision-making under uncertainty, the TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) method is integrated into the system. TOPSIS evaluates the proximity of the current userâ??s behavioral signature to an â??idealâ? and â??worst-caseâ? profile using multiple performance indicators such as typing speed, swipe dynamics, and hand stability. Each metric is weighted according to its importance in distinguishing users. This multi-criteria decision-making process ensures a balanced and fair evaluation of the userâ??s legitimacy, especially in cases where behavioral inputs may vary due to stress, fatigue, or environmental distractions. The hybrid ML-TOPSIS approach addresses common issues in behavioral biometric systems, such as intra-user variability and inter-user similarity. By combining probabilistic learning with decision prioritization, the system significantly reduces both false acceptances (unauthorized users gaining access) and false rejections (authorized users being denied). This dual-layered framework also enables adaptive learning, meaning the system updates itself over time as the userâ??s behavior subtly evolves, ensuring long-term accuracy and minimizing authentication fatigue [4].

From a security standpoint, the system is resistant to spoofing attacks, as replicating someoneâ??s behavioral pattern consistently is highly challenging. From a privacy standpoint, behavioral data do not reveal sensitive personal identifiers, which aligns with GDPR and other data protection regulations. Furthermore, this model supports continuous authentication meaning it doesnâ??t rely on one-time login events but instead continuously verifies the user in the background, enhancing security for sensitive applications like mobile banking, digital health, and secure communications. In terms of computational efficiency, the framework is optimized for mobile devices. Lightweight models and edge-computing capabilities allow real-time processing without draining battery life or requiring constant internet connectivity.TOPSIS method, being computationally simple and further aids in quick decision-making suitable for smartphone hardware limitations [5].

Conclusion

In conclusion, the hybrid use of behavioral biometrics, machine learning, and TOPSIS provides a scalable, secure, and user-friendly authentication solution. It represents a forward-thinking step in mobile security by offering a seamless and intelligent way to protect users against unauthorized access without relying solely on static credentials or invasive biometric scans. The integration of behavioral biometrics with hybrid machine learning and TOPSIS presents a powerful, user-friendly, and secure solution for smartphone authentication. This approach not only strengthens access control but also supports continuous and adaptive verification without disrupting user experience. As smartphones become central to digital identity, payment systems, and sensitive data storage, enhancing authentication through intelligent biometric frameworks is both timely and essential. With further optimization and user-specific calibration, this hybrid model can pave the way for next-generation mobile security systems that are both resilient and privacy-aware.

Acknowledgement

None.

Conflict of Interest

None.

References

1. McComb, William and V. Shanmugasundaram. "Numerical calculations of decaying isotropic turbulence using the LET theory." J Fluid Mech 143 (1984): 95–112.

Google Scholar Cross Ref

2. McNally, Anthony, John Derber, Wei Wu and Barry Katz. "The use of TOVS level-1b radiances in the NCEP SSI analysis system." Q J Roy Meteorol Soc 126 (2000): 689–724.

Google Scholar Cross Ref Indexed at

3. Ménard, Richard and Roger Daley. "The application of Kalman smoother theory to the estimation of 4DVAR error statistics." Tellus 48A (1996): 221–237.

Google Scholar Cross Ref

4. Miller, Robert, Michael Ghil and François Gauthiez. "Advanced data assimilation in strongly nonlinear dynamical systems." J Atmos Sci 8 (1994): 1037–1056.

Google Scholar Cross Ref Indexed at

5. Miller, Robert, Elizabeth Carter and Samuel Blue. "Data assimilation into nonlinear stochastic models." Tellus 51A (1999): 167–184.

Google Scholar Cross Ref