Sajjad Haider

A multi-technique approach for user identification through keystroke dynamics

Legitimate user authentication is an important part of the problems related to computer and system security. The maintenance of security becomes even more difficult when an invalid user gets the system access information. The paper presents a suite of techniques for password authentication using neural networks, fuzzy logic, statistical methods, and several hybrid combinations of these approaches. The approaches presented in the paper use typing biometrics of a user, in addition to conventional login information, to identify a user.

A heuristic approach for best sets of actions determination in influence nets

The paper presents a heuristic approach for the problem of finding the best or close-to-best sets of actions in uncertain situations modeled by influence nets. The approach enhances the analysis capability of influence nets by allowing the user to observe the combined impact of actions on the desired effect in contrast to the sensitivity analysis that allows the user to evaluate individual impacts only. Unlike the exhaustive search which works in exponential time, the proposed approach generates result in polynomial time. The paper also demonstrates the generalization of alternative sets of actions.

Effective Course-of-Action Determination to Achieve Desired Effects

An evolutionary algorithm-based approach to identify effective courses of action (COAs) in dynamic uncertain situations is presented. The uncertain situation is modeled using timed influence nets, an instance of dynamic Bayesian networks. The approach makes significant enhancements to the current trial-and-error-based manual technique, which is not only labor intensive but also not capable of modeling constraints among actionable events. The proposed approach is an attempt to overcome these limitations. It automates the process of COA identification. It also allows a system analyst to capture certain types of constraints among actionable events. Because of its parallel search nature, the approach produces multiple COAs that have a similar fitness value. This feature not only gives more flexibility to a decision maker during mission planning, but it can also be used to generalize the COAs if there exists a pattern among them. This paper also discusses a heuristic that further enhances the performance of the approach.

Suspicious activity reporting using dynamic bayesian networks

Abstract Suspicious activity reporting has been a crucial part of anti-money laundering systems. Financial transactions are considered suspicious when they deviate from the regular behavior of their customers. Money launderers pay special attention to keep their transactions as normal as possible to disguise their illicit nature. This may deceive the classical deviation based statistical methods for finding anomalies. This study presents an approach, called SARDBN (Suspicious Activity Reporting using Dynamic Bayesian Network), that employs a combination of clustering and dynamic Bayesian network (DBN) to identify anomalies in sequence of transactions. SARDBN applies DBN to capture patterns in a customer’s monthly transactional sequences as well as to compute an anomaly index called AIRE (Anomaly Index using Rank and Entropy). AIRE measures the degree of anomaly in a transaction and is compared against a pre-defined threshold to mark the transaction as normal or suspicious. The presented approach is tested on a real dataset of more than 8 million banking transactions and has shown promising results.

BNOSA: A Bayesian network and ontology based semantic annotation framework

Abstract The paper presents a semantic annotation framework that is capable of extracting relevant information from unstructured, ungrammatical and incoherent data sources. The framework, named BNOSA, uses ontology to conceptualize a problem domain and to extract data from the given corpora, and Bayesian networks to resolve conflicts and to predict missing data. The framework is extensible as it is capable of dynamically extracting data from any problem domain given a pre-defined ontology and a corresponding Bayesian network. Experiments have been conducted to analyze the performance of BNOSA on several problem domains. The sets of corpora used in the experiments belong to selling–purchasing websites where product information is entered by ordinary web users in a structure-free format. The results show that BNOSA performs reasonably well to find location of the data of interest using context keywords provided as part of the domain ontology. In case of more than one value being extracted for an attribute or if the value is missing, Bayesian networks identify the most appropriate value for that attribute.

Modeling time-varying uncertain situations using Dynamic Influence Nets

This paper enhances the Timed Influence Nets (TIN) based formalism to model uncertainty in dynamic situations. The enhancements enable a system modeler to specify persistence and time-varying influences in a dynamic situation that the existing TIN fails to capture. The new class of models is named Dynamic Influence Nets (DIN). Both TIN and DIN provide an alternative easy-to-read and compact representation to several time-based probabilistic reasoning paradigms including Dynamic Bayesian Networks. The Influence Net (IN) based approach has its origin in the Discrete Event Systems modeling. The time delays on arcs and nodes represent the communication and processing delays, respectively, while the changes in the probability of an event at different time instants capture the uncertainty associated with the occurrence of the event over a period of time.

Efforts to blend ontology with Bayesian networks: An overview

The concept of ontology modeling and engineering has drastically evolved with the emergence of semantic web. Today in a collaborative environment ontologies need to interoperate for efficient data exchange. Due to their deterministic nature, ontologies are unable to capture uncertainty inherent in many real world problems. Efforts have been made to add the dimension of uncertainty by synergizing Bayesian network with ontology - a combination in which domain knowledge and probabilistic information go hand in hand. This paper categorizes different approaches that have been proposed to blend ontology and Bayesian networks. It also discusses the scope of these approaches and their intended application areas. Their relative strengths and weaknesses are also highlighted. The paper contributes significantly by providing a comprehensive comparison which could serve as the starting point for new researchers.

Ontology Based Expert-System for Suspicious Transactions Detection

The development of an effective mechanism to detect suspicious transactions is a critical problem for financial institutions in their endeavor to prevent anti-money laundering activities. This research addresses this problem by proposing an ontology based expert-system for suspicious transaction detection. The ontology consists of domain knowledge and a set of (SWRL) rules that together constitute an expert system. The native reasoning support in ontology is used to deduce new knowledge from the predefined rules about suspicious transactions. The presented expert-system has been tested on a real data set of more than 8 million transactions of a commercial bank. The novelty of the approach lies in the use of ontology driven technique that not only minimizes the data modeling cost but also makes the expert-system extendable and reusable for different applications.

Clustering based anomalous transaction reporting

Abstract Anti-money laundering (AML) refers to a set of financial and technological controls that aim to combat the entrance of dirty money into financial systems. A robust AML system must be able to automatically detect any unusual/anomalous financial transactions committed by a customer. The paper presents a hybrid anomaly detection approach that employs clustering to establish customers’ normal behaviors and uses statistical techniques to determine deviation of a particular transaction from the corresponding group behavior. The approach implements a variant of Euclidean Adaptive Resonance Theory, termed as TEART, to group customers in different clusters. The paper also suggests an anomaly index, named AICAF, for ranking transactions as anomalous. The approach has been tested on a real data set comprising of 8.2 million transactions and the results suggest that TEART scales well in terms of the partitions obtained when compared to the traditional K-means algorithm. The presented approach marks transactions having high AICAF values as suspicious.

Teaching coordinated strategies to soccer robots via imitation

Developing coordination among multiple agents and enabling them to exhibit teamwork is a challenging yet exciting task that can benefit many of the complex real-life problems. This research uses imitation to learn collaborative strategies for a team of agents. Imitation based learning involves learning from an expert by observing him/her demonstrating a task and then replicating it. The key idea is to involve multiple human experts during demonstration to teach autonomous agents how to work in coordination. The effectiveness of the proposed methodology has been assessed in a goal defending scenario of the RoboCup Soccer Simulation 3D league. The process involves multiple human demonstrators controlling soccer agents via game controllers and demonstrating them how to play soccer in coordination. The data gathered during this phase is used as training data to learn a classification model which is later used by the soccer agents to make autonomous decisions during actual matches. Different performance evaluation metrics are derived to compare the performance of imitating agent with that of the human-driven agent and hand-coded (if-then-else rules) agent.