Sangeeta Mittal

Situation recognition in sensor based environments using concept lattices

A variety of sensors are available nowadays for fine grain continuous monitoring of our environments in many desired ways. Comprehension of streams of data from deployed sensors in a meaningful way is critical to usability of the sensors. One such comprehension is context and situation which may affect our actions and decisions. Context is deduced from the sensor data using probabilistic methods like maximum likelihood estimation and Bayesian probabilities. Possible situations are abstracted using deduced contexts. Event trees, template and rule based methods have been used for deriving situations from contexts. Lattices are constructed using formal concept analysis methods for representation and recognition of situations. When the context information is either noisy or incomplete, set of Implication and Association Rules are derived from the lattice and used for situation recognition. For illustration, Situation of an elderly person living alone in a house, deployed with various sensors is recognized using the above technique.

A review of some Bayesian Belief Network structure learning algorithms

Bayesian Belief Networks (BBNs) are useful in modeling complex situations. Such graphical models help in giving better insight and understanding of the situation. Many algorithms for machine learning of BBN structures have been developed. In this paper six different algorithms have been reviewed by constructing BBN structures for two different datasets using various algorithms. Some inferences have been drawn from the results obtained from the study which may help in decision making.

Preprocessing methods for context extraction from multivariate wireless sensors data — An evaluation

Real time extraction of user context like current location, position, physiological parameters, actions etc. from sensor data is a challenging problem. Probabilistic machine learning algorithms or their variations are used as classifier for accomplishing this task. Sensor data in raw form is voluminous, redundant, noisy and real valued. Most classification schemes cannot perform well on such data. Raw data is preprocessed to counter these inherent problems and transform the data to become more suitable for context extraction algorithms. Horizontal and vertical data reductions are main preprocessing methods. In this work, these methods of data reduction are emphasized and studied. These include imputation, smoothing, feature extraction and data reduction in sequential order. Representative methods applied in conventional data mining are evaluated. A benchmark sensor dataset of ambient, object and wearable sensors is used for the study. Data apart from being real valued also has scores of missing values. The “Mode of Locomotion” context of observed person is extracted by applying Bayesian Belief Network based classifier post processing. Horizontal data reduction has not found to be suitable for sensor data.

Effect of Choice of Discretization Methods on Context Extraction from Sensor Data --- An Empirical Evaluation

Data from sensor logs in raw form is generally continuous valued. This data from multiple sensors in continuous stream becomes voluminous. For knowledge discovery like extraction of context, from these datasets, standard machine learning algorithms or their variations are used as classifier. Most classification schemes require the input data to be discretized. The focus of this paper is to study merits of some popular discretization methods when applied on noisy sensor logs. Representative methods from supervised and unsupervised discretization, like binning, clustering and entropy minimization are evaluated with context extraction. Interestingly, unlike common perception, for discretization of sensor data, supervised algorithms do not have a clear edge over unsupervised.

A Novel Bayesian Belief Network Structure Learning Algorithm Based on Bio-Inspired Monkey Search Meta Heuristic

Bayesian Belief Networks (BBN) combine available statistics and expert knowledge to provide a succinct representation of domain knowledge under uncertainty. Learning BBN structure from data is an NP hard problem due to enormity of search space. In recent past, heuristics based methods have simplified the search space to find optimal BBN structure (based on certain scores) in reasonable time. However, slow convergence and suboptimal solutions are common problems with these methods. In this paper, a novel searching algorithm based on bio-inspired monkey search meta-heuristic has been proposed. The jump, watch-jump and somersault sub processes are designed to give a global optimal solution with fast convergence. The proposed method, Monkey Search Structure Leaner (MS2L), is evaluated against five popular BBN structure learning approaches on model construction time and classification accuracy. The results obtained prove the superiority of our proposed algorithm on all metrics.

iABC: Towards a hybrid framework for analyzing and classifying behaviour of iOS applications using static and dynamic analysis

Abstract Is this app safe to use? - A wrong decision can result in privacy breach in iOS devices. In this digital era users extensively use smart devices to store their personal and important information. To ease users’ tasks, thousands of free or paid apps are available in app store. However, recent studies reveal startling facts about various attacks and data harvesting incidents through these apps, where personal data is put at risk. Through this paper, we propose a permission induced risk model- iOS Application analyzer and Behavior Classifier (iABC), for iOS devices to detect privacy violations arising due to granting permissions during installation of applications. It is a two-layer process comprising of static and dynamic analysis. It uses reverse engineering to extract permission variables from applications and computes a risk score for each application using ranking algorithms. The approach considers applications category as a key feature for detecting malicious applications while computing static risk score. Different machine learning classifiers were employed to evaluate 1,150 applications. The empirical results show that our proposed model gives detection rate of 97.04%. Furthermore, to assess privacy breaches by applications at run time, dynamic analysis on 50 applications has been performed to obtain dynamic risk scores of installed apps.

Performance Evaluation of Openflow SDN Controllers

Software Defined Networks (SDN) is the recent networking paradigm being adopted by stakeholders in big way. The concept works towards dramatically reducing network deployment and management costs. Controllers, also known as Network Operating System of SDN, are critical to success of SDNs. Many open source controllers are available for use. In this paper, performance of four popular Openflow based controllers has been evaluated on various metrics. Latency, Bandwidth utilization, Packet Transmission rate, Jitter and Packet loss has been calculated for TCP and UDP traffic on varying network sizes, topologies and Controllers. Floodlight is one of the best performing as compared to reference controller.

Unmasking non-simultaneous sybils in mobile opportunistic networks

Mobile opportunistic networks allow nodes to freely join and leave the network; hence they are highly susceptible to security threats. The paper proposes a method for detection and avoidance of a non-simultaneous Sybil attack in mobile opportunistic networks. In this attack, a malicious node assumes multiple, fake identities (i.e. Sybils) over a period of time to remain active in the network even after it has been detected as malicious. These Sybils may deplete the network resources or bias recommendation and voting systems. The proposed Sybil detection and avoidance methodology has been simulated using the opportunistic network simulator, ONE. The simulation results show that the proposed technique achieves a higher probability of packet delivery and reduces the average latency, overhead and the number of aborted message transmissions in networks with varying number of malicious nodes as compared to the case when no Sybil detection method is used.