Description:TITLE OF THE INVENTION: AI-BASED BEHAVIORAL BIOMETRIC AUTHENTICATION FOR E-COMMERCE
FIELD OF THE INVENTION
[0001] The present invention relates to the field of Cybersecurity and Artificial Intelligence. More particularly, the present invention relates to a system for AI-based behavioral biometric authentication for e-commerce.
BACKGROUND OF THE INVENTION
[0002] The field of e-commerce has expanded rapidly, leading to an increasing need for secure and reliable authentication methods to protect user accounts and financial transactions. Traditional authentication approaches, such as passwords, PINs and one-time passcodes, are vulnerable to security threats like phishing, credential theft and brute-force attacks. Biometric authentication methods, including fingerprint and facial recognition, have improved security but still present challenges related to spoofing, data breaches and privacy concerns. Additionally, hardware-based biometrics require specialized sensors, limiting their accessibility across all devices and platforms.
[0003] Behavioral biometrics have emerged as an alternative authentication method that relies on analyzing user interactions, such as keystroke dynamics, mouse movements and touchscreen gestures. Traditional behavioral authentication systems often use rule-based or statistical models to identify user patterns. However, these approaches struggle with adapting to variations in user behavior over time and may generate high false acceptance or rejection rates. Moreover, existing methods cannot continuously authenticate users in real time, making them less effective against sophisticated cyber threats.
[0004] So, the present invention provides an AI-based behavioral biometric authentication for e-commerce by analyzing user interactions such as typing speed, mouse dynamics, swipe gestures and touch pressure. Behavioral data is captured and pre-processed to extract key features. An LSTM model analyzes these features to detect patterns and identify legitimate users. A decision tree-based anomaly detection module classifies behavior as normal or suspicious. If anomalies are detected, fraud alerts are triggered to prevent unauthorized access. User profiles are continuously updated to adapt to behavioral changes, ensuring secure and reliable e-commerce transactions.
[0005] However, there were lots of efforts made to authenticate users in e-commerce transactions. Some of the references known to us are as follows:
Prior Arts:
[0006] US9531710B2 describes a system and method of a behavioral authentication system using a biometric fingerprint sensor and user behavior for authentication. The device integrates a biometric input device (e.g., fingerprint sensor, facial recognition, or voice recording) and a behavioral input device (e.g., touchpad, keystroke sensor, gyroscope, accelerometer, and network activity) to create user profiles. A processor generates and stores biometric and behavioral user profiles in a database, later comparing them with real-time samples during authentication. The system analyzes fingerprint swipes (direction, speed, pressure, and position) and behavioral traits (device orientation, location, network activity, app usage, and usage time patterns). A numeric security classification determines user access levels based on the degree of correlation between stored and real-time data. The authentication process can be performed locally or via a remote server.
[0007] US10210318B2 describes the system and method of methods and systems for capturing biometric data using a terminal device equipped with an accelerometer, gyroscope, processor, and display. The device determines its orientation by calculating the angle between its surface and the X-axis based on motion data. If the angle deviates from 90 degrees, the device guides the user into an optimal position by displaying an image of a biometric modality (e.g., face) along with two pairs of parallel lines. These lines move closer as the user adjusts the device, ensuring proper alignment for biometric capture. Once the lines overlap, the device automatically captures biometric data, improving accuracy and trustworthiness in authentication transactions. The system further enhances reliability by ensuring the device is at the user's face level before capturing the biometric sample which improves biometric authentication by minimizing errors due to improper positioning.
[0008] US8739278B2 describes a system and method for fraud monitoring and detection using application fingerprinting, which involves generating unique identifiers, or fingerprints, based on data submitted by users within a software application. The invention detects anomalous or potentially fraudulent submissions by comparing a newly generated fingerprint, associated with a user's input and contextual information (such as device, location, and workflow), with previously recorded fingerprints of legitimate users. A risk score is calculated based on the degree of similarity between the new and historical fingerprints, with additional factors like blacklisted devices or locations influencing the score. If the risk score exceeds a threshold, the system may trigger authentication challenges or alerts for administrators. The technology can be applied in web-based applications to enhance security and prevent malicious activities such as automated attacks or unauthorized access attempts.
[0009] US20180101831A1 describes the system and method for performing secure online banking transactions by dynamically verifying the security of a banking operation before execution. The invention involves collecting transaction-related data, generating multiple verification scripts with different rules for gathering identification data and executing these scripts to assess security risks. The collected data, which may include device logs, RAM dumps, and network activity, is categorized and analyzed using numerical security coefficients to compute an overall security level. If the security level meets or exceeds a predefined threshold, the banking operation is approved; otherwise, it may be blocked or subjected to further authentication. This approach enhances fraud prevention by adapting verification processes based on transaction context and potential threats.
[0010] State-of-the-art suffers from the following limitations:
[0011] The state of the arts does not consider AI-based behavioral biometric authentication for e-commerce. Existing methods primarily rely on static credentials, such as passwords or PINs, and enhanced security measures like two-factor authentication (2FA), which fail to detect behavioral anomalies and adapt to evolving fraud techniques. So, the present invention provides AI-based behavioral biometric authentication for e-commerce that utilizes user interactions, including typing speed, mouse dynamics, swipe gestures and touch pressure. Behavioral data is captured during user transactions and processed through a data acquisition module, which extracts unique interaction patterns. This data is then analyzed by an LSTM network to recognize and retain behavioral dependencies, ensuring accurate user verification. An anomaly detection mechanism classifies the behavior as normal or suspicious, triggering fraud alerts when necessary. The authentication results are used to approve or decline transactions, ensuring secure e-commerce interactions while continuously updating user profiles to adapt to behavioral changes.
OBJECTIVES OF THE INVENTION
[0012] The main objective of the present invention is to provide AI-based behavioral biometric authentication for e-commerce by analyzing user interactions to ensure secure transaction approvals.
[0013] Another objective of the present invention is to capture behavioral biometric data, including typing speed, mouse dynamics, swipe gestures and touch pressure, during user interactions.
[0014] Still, another objective of the present invention is to employ Long Short-Term Memory (LSTM) networks to analyze behavioral data, recognize patterns and detect anomalies in user authentication.
[0015] Another objective of the present invention is to classify fraudulent and legitimate user activities by analyzing behavioral patterns using decision trees, which evaluate multiple behavioral attributes and identify anomalies based on predefined classification criteria.
[0016] Further objective of the present invention is to continuously update user behavioral profiles, adapting to legitimate changes while maintaining secure authentication for e-commerce transactions.
SUMMARY OF THE INVENTION
[0017] The present invention summary is easy to understand before the hardware and system enablement were illustrated in this present invention. There have been multiple possible embodiments that do not expressly point up in this method's present acknowledgment. Here, the conditions are used to explain the purpose of exacting versions or embodiments for understanding the present invention
[0018] The main aspect of the present invention is AI-based behavioral biometric authentication for e-commerce, which involves analyzing user interactions, including keystroke dynamics, mouse movements, and browsing patterns, to authenticate users based on behavioral biometrics. The process begins with the continuous collection of user interaction data, ensuring accurate profiling for authentication. This data is processed through an AI- module, where behavioral attributes are extracted and compared against a reference profile. Decision trees are employed to classify user behaviors by evaluating multiple attributes and determining whether an action aligns with the expected pattern. When deviations are detected, an alert is triggered to prevent fraudulent access.
[0019] Another aspect of the present invention is the collection of behavioral data, including keystroke timing, mouse trajectory and click intensity, to establish a unique user profile to make sure that the authentication is based not only on credentials but also on behavioral consistency, reducing the risk of identity theft. The collected data is analyzed in real-time to detect inconsistencies and provide continuous authentication throughout the session.
[0020] Yet another aspect of the present invention is the use of Long Short-Term Memory (LSTM) networks for sequential analysis of user behavior over time. LSTM processes time-series behavioral data, such as typing rhythm changes and navigation patterns, to identify long-term trends in user interactions which allows for adaptive authentication, where the system continuously learns and refines user profiles, improving accuracy in fraud detection and reducing false positives.
[0021] Another aspect of the present invention is the decision tree-based classification, which is utilized to evaluate behavioral attributes at multiple levels. Decision trees systematically assess various parameters, such as typing speed variations, cursor movement deviations, and response latencies, to determine authentication confidence. By constructing hierarchical decision nodes, the system differentiates between legitimate users and potentially fraudulent activities based on historical behavioral patterns.
[0022] Yet another aspect of the present invention is the AI module to detect anomalies by continuously monitoring user behavior and comparing it with historical data from past sessions. When the system identifies a significant deviation from the established behavioral profile, it triggers a multi-factor authentication request or restricts access to mitigate potential fraud risks. The decision trees within the AI module enhance anomaly detection by adjusting classification thresholds based on user interactions.
[0023] Still, another aspect of the present invention is the generation of authentication reports that provide details into user behavior trends, login success rates and detected anomalies. These reports include graphical representations that highlight authentication attempts, suspicious activities and session-based behavior metrics, helping e-commerce platforms improve security measures and fraud prevention strategies.
BRIEF DESCRIPTION OF THE DRAWINGS
[0024] The accompanying drawings, which are incorporated, constitute a part of the specification, illustrate the invention's embodiment, and the description serves to explain the principles of the invention.
[0025] Various embodiments will be described under the appended drawings, which are provided to illustrate the present invention
[0026] Figure 1 illustrates the system architecture of the present invention as provided in the present invention.
[0027] Figure 2 illustrates the flow chart of the present invention as provided in the present invention.
[0028] Figure 3 illustrates the workflow process of the present invention as provided in the present invention.
[0029] Figure 4 illustrates the decision tree model for fraud detection as provided in the present invention
DETAILED DESCRIPTION OF THE INVENTION
[0030] The present invention is easily understood with references, detailed descriptions, block diagrams, and figures. Here, various embodiments have been discussed regarding the block diagram, architecture, and other references. Some embodiments of this invention, illustrating its features, will now be discussed, and the disclosed embodiments are merely exemplary of the invention that may be embodied in various forms.
[0031] Traditional authentication methods for e-commerce transactions primarily rely on static credentials such as passwords, PINs, and security questions. While these methods are widely used, they have several limitations, including vulnerability to phishing attacks, password theft, and unauthorized access due to weak or reused credentials. Two-factor authentication (2FA) has been introduced to enhance security, requiring users to provide additional verification, such as a one-time password (OTP) or biometric scan. However, these methods still fall short in detecting subtle behavioral anomalies and adapting to evolving user behavior, leaving gaps in preventing advanced fraud techniques like account takeover or identity theft. As a result, there is a growing need for more robust, behavior-driven authentication solutions to improve security in e-commerce transactions.
[0032] So, the present invention provides an AI-based behavioral biometric authentication system for e-commerce that authenticates users based on their interaction patterns, such as keystroke dynamics, mouse movements and browsing behavior. Unlike traditional authentication methods that rely solely on passwords or one-time codes, this approach continuously monitors user behavior throughout their session, creating an authentication process. By utilizing behavioral biometrics, unauthorized access attempts can be detected in real time, even if correct login credentials are used to ensure a higher level of security for e-commerce transactions while minimizing friction for legitimate users.
[0033] The process collects and examines multiple behavioral attributes, such as typing speed variations, cursor movements, scrolling patterns and response times. These behavioral data points are processed using AI models like decision trees and Long Short-Term Memory (LSTM) networks to identify patterns unique to each user. Decision trees classify behavior based on set thresholds, distinguishing between typical and unusual activities. LSTM networks analyze sequential behavioral data over time, helping AI modules recognize long-term patterns and adapt to changes in user behavior, reducing false positives and enhancing reliability.
[0034] To enhance fraud detection, the present invention continuously compares real-time interactions with the user's historical behavior profile. Any significant deviation, such as erratic cursor movement, unfamiliar typing rhythm or unusual navigation sequences, triggers an anomaly detection mechanism. If an anomaly is detected, the system prompts additional verification steps, such as multi-factor authentication or temporary session lockdown, to prevent unauthorized transactions.
[0035] Additionally, the invention generates detailed authentication reports that summarize user activity trends, login attempts and detected anomalies. These reports include visual analytics, such as graphs and statistical breakdowns, to help e-commerce platforms monitor security incidents and refine fraud prevention strategies. By utilizing AI-based behavioral biometric authentication, e-commerce platforms can enhance transaction security while providing a user experience, reducing reliance on static credentials and improving fraud prevention measures.
[0036] The present invention shows the perspective view of student and teacher behavior analysis using a CNN (100) further detailed descriptions of the present invention are stated here in the attached drawings. Thus, the detailed embodiments of the present invention are disclosed here to describe the present invention.
[0037] In this embodiment of the present invention, as shown in the figure.1 refers to the system architecture of the present invention which comprises data acquisition (101), Pre-processing (102), AI model (103), Anomaly detection (104) and feedback (105).
[0038] An AI-based behavioral biometric authentication system for e-commerce, which uses unique user behavioral patterns, such as typing speed, mouse dynamics, swipe gestures and touch pressure, to ensure secure and reliable transaction approvals, which comprises a Data Acquisition (101), Pre-processing (102), AI Model (103), Anomaly Detection (104), and Feedback (105). The process starts with Data Acquisition (101), where various user behavioral biometrics are collected, including Typing Speed (1011), Mouse Dynamics (1012), Swipe Gestures (1013), and Touch Pressure (1014). These inputs form a unique behavioral profile for each user based on their interaction patterns. Typing speed captures keystroke timing variations, mouse dynamics monitor movement behavior, swipe gestures track navigation habits and touch pressure assesses screen interaction strength.
[0039] Once the data is gathered, it undergoes Pre-processing (102) to refine and structure it for further analysis. Different techniques are applied based on data type Noise Removal (1021) refines typing speed data, Data Normalization (1022) adjusts variations in mouse dynamics, Feature Extraction (1023) isolates key swipe gesture patterns and Segmentation (1024) processes touch pressure data for better analysis to ensure that the AI model receives structured and optimized data, improving recognition and authentication performance.
[0040] The AI Model (103) processes the pre-processed data to learn and analyze user behavior. It utilizes Long Short-Term Memory (LSTM) networks (1031) to understand sequential dependencies, making it possible to detect subtle behavior changes. Additionally, Behavioral Pattern Analysis (1032) establishes a baseline of normal user behavior by continuously evaluating interactions. If the behavioral input aligns with the user’s profile, authentication proceeds smoothly. However, any deviation raises suspicion, prompting further examination.
[0041] To detect anomalies, Anomaly Detection (104) methods such as Decision Trees (1041) and Threshold Analysis (1042). Decision Trees classify behavioral patterns based on predefined logical rules, while Threshold Analysis identifies actions that exceed or fall below acceptable limits. If an anomaly is detected, the system initiates the Feedback (105) process. This includes Profile Updates (1051) to refine user models, Fraud Alerts (1052) to notify security teams of potential threats and User Notifications (1053) requesting additional verification steps.
[0042] The other embodiment of the present invention is shown in Figure 2, which refers to the flow chart (200) outlining the complete process of behavioral biometric authentication, from data capture and preprocessing to model analysis, comparison and user authentication or alert generation.
[0043] Figure 2 presents a flowchart illustrating the process flow of the AI-based behavioral biometric authentication system. The process begins with capturing user behavioral data. This data is then checked for validity. If the data is deemed invalid, the system requests new data, ensuring the input is suitable for processing. Upon validation, the data undergoes preprocessing, which prepares it for feature extraction. Features that are essential for behavioral analysis are extracted, highlighting the unique characteristics of the user's interaction patterns. These extracted features are then fed into the Long Short-Term Memory (LSTM) model, which plays a central role in analyzing the user's behavior over time.
[0044] The LSTM model analyzes the behavioral patterns and compares them with the stored profile of the user which is an essential step in determining whether the current behavior aligns with the user's typical interaction style. If the pattern matches the stored profile ("Yes"), the user is authenticated, granting them access. Simultaneously, the system proceeds to update the user profile, if needed, ensuring the profile remains current and reflective of the user's evolving behavior which is essential for maintaining the system's reliability and minimizing false positives.
[0045] Conversely, if the pattern does not match the stored profile ("No"), the system generates an alert, indicating an unauthorized access attempt. This alert triggers further investigation or security measures, such as multi-factor authentication or account lockdown, to protect the user and the system. The flowchart concludes after either authentication or alert generation, representing the completion of a single authentication cycle.
[0046] The other embodiment of the present invention is shown in Figure 3, which refers to the workflow process of the LSTM (300) outlining the complete sequence of operations within the LSTM cell, from input processing and gate management to cell state updates and the generation of a behavioral representation for authentication and anomaly detection.
[0047] Figure 3 illustrates the workflow process of the Long Short-Term Memory (LSTM) network within the AI-based behavioral biometric authentication system. The process begins with receiving sequential input data, encompassing various behavioral biometrics such as typing speed, mouse dynamics, and swipe gestures. This raw data is then fed into the LSTM cell, which is initialized to begin processing the sequence. The LSTM cell's architecture is key to its functionality, featuring three primary gates: the Forget Gate, the Input Gate, and the Output Gate.
[0048] The Forget Gate plays an essential role in managing information from previous time steps. It decides which past information to retain for future analysis and which information to discard as irrelevant. This selective memory is essential for the LSTM to focus on the most pertinent aspects of the user's behavioral patterns. Simultaneously, the Input Gate determines what new information from the current input should be stored in the cell state. This gate regulates the addition of new data into the LSTM's long-term memory, ensuring that only relevant information is incorporated.
[0049] The cell state update is where the core learning occurs. The LSTM updates its long-term memory by adding the new relevant information identified by the Input Gate and removing irrelevant information as dictated by the Forget Gate. This updating process allows the LSTM to adapt and learn from the evolving sequence of behavioral data. Finally, the Output Gate generates the final behavioral representation for authentication which encapsulates the learned patterns and is used for comparison with the user's stored profile.
[0050] The processed data is then passed to a Decision Tree for anomaly detection. If the behavior is deemed normal, the transaction is approved and the user is authenticated. However, if the behavior is anomalous, potentially indicating fraudulent activity, a fraud alert is triggered, prompting further investigation or security measures. The process concludes after either approval or alert generation, marking the completion of a single authentication cycle within the LSTM behavioral biometric system which enables the system to learn complex temporal patterns in user behavior, contributing to robust and adaptive authentication.
[0051] The other embodiment of the present invention shown in Figure 4 refers to the decision tree model for fraud detection.
[0052] Figure 4 depicts the Decision Tree Model employed for fraud detection within the AI-based behavioral biometric authentication system. The process begins with the Decision Tree Generation which utilizes training data, specifically the processed behavioral data received from the LSTM network. The decision tree method selects the best attributes for decision-making using Attribute Selection Measures (ASM) such as Information Gain, Gini Index or Gain Ratio. These metrics evaluate the effectiveness of different attributes in classifying the data, ensuring the most informative attributes are chosen for the tree's nodes.
[0053] Once the best attributes are selected, the dataset is divided into smaller subsets based on the values of these attributes. This process creates branches in the decision tree. This process is then recursively repeated for each child node or subset, progressively building the tree structure. This recursive partitioning continues until a stopping criterion is met, such as a maximum tree depth or a minimum number of samples in a leaf node. The result is a hierarchical tree-like structure where each node represents a decision based on an attribute, each branch represents an outcome of the decision, and each leaf node represents a final classification.
[0054] Following the tree generation, the model undergoes "Model Evaluation." This step involves using a separate set of data, the "Testing Data," which was not used in the training phase. The testing data is passed through the decision tree and the model's classifications are compared to the actual known classifications of the testing data. This evaluation provides details into the model's performance metrics, such as accuracy, precision, recall and F1-score. These metrics are essential for assessing the model's ability to generalize to unseen data and its overall effectiveness in fraud detection.
[0055] The output of the Decision Tree Model is a "Classified Result," which categorizes the input behavior as either "Legitimate" or "Fraudulent." If the behavior aligns with the user's established patterns, it is classified as legitimate, allowing the transaction to proceed. Conversely, if the behavior deviates significantly and triggers the model's anomaly detection rules, it is flagged as fraudulent, prompting further investigation or security measures. This classification provides a pivotal layer of real-time security, enabling the system to automatically identify and respond to potential threats in e-commerce transactions. , Claims:We claim,
1. An AI-based behavioral biometric authentication system (100) for e-commerce, designed to authenticate users based on unique behavioral patterns, comprising: (i) a Data Acquisition Unit (101) that collects behavioral biometrics, including typing dynamics, mouse movements, and touchscreen interactions; (ii) a Preprocessing Unit (102) that refines collected data by removing noise and normalizing variations; (iii) a Behavioral Feature Extraction Unit (103) that extracts key behavioral traits for authentication; (iv) an AI-based Classification Model (104) utilizing machine learning techniques to analyze behavioral patterns; (v) an Anomaly Detection Unit (105) that identifies deviations in user behavior to detect fraudulent activities; and (vi) a Security Response Unit (106) that generates authentication decisions, alerts, and prompts for additional verification, characterized in that:
a) a Long Short-Term Memory (LSTM) Network (1041) processes sequential behavioral data, such as typing dynamics and mouse movements, by capturing temporal dependencies. It learns patterns over time, distinguishing between normal and anomalous user behavior; and
b) a Decision Tree Model (1042) classifies behavioral variations by applying structured decision rules to user interaction patterns. It segments data based on key features like typing dynamics, swipe gestures, and mouse movements to distinguish between normal and suspicious behavior. Each decision node evaluates specific characteristics, leading to an authentication or alert outcome. This model ensures fast and interpretable classification of behavioral anomalies.
2. The AI-based behavioral biometric authentication system (100) as claimed in Claim 1, wherein the said Preprocessing Unit (102) applies Noise Filtering (1021) to remove inconsistencies from typing and touch data. Data Normalization (1022) to standardize variations in mouse movement patterns. Feature Scaling (1023) to refine extracted behavioral attributes for analysis.