Alistair Shilton

Detecting Selective Forwarding Attacks in Wireless Sensor Networks using Support Vector Machines

Wireless sensor networks (WSNs) are a new technology foreseen to be used increasingly in the near future due to their data acquisition and data processing abilities. Security for WSNs is an area that needs to be considered in order to protect the functionality of these networks, the data they convey and the location of their members. The security models and protocols used in wired and other networks are not suited to WSNs because of their severe resource constraints, especially concerning energy . In this article, we propose a centralized intrusion detection scheme based on support vector machines (SVMs) and sliding windows. We find that our system can detect black hole attacks and selective forwarding attacks with high accuracy without depleting the nodes of their energy.

Distributed data fusion using support vector machines

The basic quantity to be estimated in the Bayesian approach to data fusion is the conditional probability density function (CPDF). Computationally efficient particle filtering approaches are becoming more important in estimating these CPDFs. In this approach, IID samples are used to represent the conditional probability densities. However, their application in data fusion is severely limited due to the fact that the information is stored in the form of a large set of samples. In all practical data fusion systems that have limited communication bandwidth, broadcasting this probabilistic information, available as a set of samples, to the fusion center is impractical. Support vector machines, through statistical learning theory, provide a way of compressing information by generating optimal kernal based representations. In this paper we use SVM to compress the probabilistic information available in the form of IID samples and apply it to solve the Bayesian data fusion problem. We demonstrate this technique on a multi-sensor tracking example.

A Division Algebraic Framework for Multidimensional Support Vector Regression

In this paper, division algebras are proposed as an elegant basis upon which to extend support vector regression (SVR) to multidimensional targets. Using this framework, a multitarget SVR called ¿Z-SVR is proposed based on an ¿-insensitive loss function that is independent of the coordinate system or basis used. This is developed to dual form in a manner that is analogous to the standard ¿-SVR. The ¿H-SVR is compared and contrasted with the least-square SVR (LS-SVR), the Clifford SVR (C-SVR), and the multidimensional SVR (M-SVR). Three practical applications are considered: namely, 1) approximation of a complex-valued function; 2) chaotic time-series prediction in 3-D; and 3) communication channel equalization. Results show that the ¿H-SVR performs significantly better than the C-SVR, the LS-SVR, and the M-SVR in terms of mean-squared error, outlier sensitivity, and support vector sparsity.

Adaptive support vector machines for regression

Support vector machines are a general formulation for machine learning. It has been shown to perform extremely well for a number of problems in classification and regression. However, in many difficult problems, the system dynamics may change with time and the resulting new information arriving incrementally will provide additional data. At present, there is limited work to cope with the computational demands of modeling time varying systems. Therefore, we develop the concept of adaptive support vector machines that can learn from incremental data. Results are provided to demonstrate the applicability of the adaptive support vector machines techniques for pattern classification and regression problems.

Distributed training of multiclass conic-segmentation support vector machines on communication constrained networks

We present a distributed algorithm for training multiclass conic-segmentation support vector machines (CS-SVMs) on communication-constrained networks. The proposed algorithm takes advantage of the sparsity of the CS-SVM to minimise the communication overhead between nodes during training to obtain classifiers at each node which closely approximate the optimal (centralised) classifier. The proposed algorithm is also suited for wireless sensor networks where inter-node communication is limited by power restrictions and bandwidth. We demonstrate our algorithm by applying it to two datasets, one simulated and one benchmark dataset, to show that the global decision functions found by the nodes closely approximate the optimal decision function found by a centralised algorithm possessing all training data in one batch.

Combined multiclass classification and anomaly detection for large-scale Wireless Sensor Networks

A smart wireless sensor network analytics requirement, beyond routine data collection, aggregation and analysis, in large-scale applications, is the automatic classification of emerging unknown events (classes) from the known classes. In this paper we present a new form of SVM that combines multiclass classification and anomaly detection into a single step to improve performance when data contains vectors from classes not represented in the training set. We demonstrate how the concepts of structural risk minimisation and anomaly detection are combined and analysing the effect of the various training parameters. The evaluations on several benchmark datasets reveal its ability to accurately classify unknown classes and known classes simultaneously.

DP1SVM: A dynamic planar one-class support vector machine for Internet of Things environment

The Internet of Things realisations, such as smart city applications, generates a vast amount of data, and detecting emerging anomalies in such large unlabelled data is a challenge. One-class support vector machines (1SVMs) have ability to detect anomalies by modelling the complex normal patterns in the data. However, they have limitations in terms of higher time complexity. Dynamically updating the 1SVM model for a streaming data by retraining from scratch is a time consuming task. In this work we present a dynamic planar 1SVM that can not only incrementally learn new data as well as remove historic data decrement-ally from the system, but also dynamically adjust the parameters of the algorithm. Evaluation on simulated and benchmark datasets reveals its ability to effectively re-learn with significantly lower computational overhead. Moreover, we analyse its performance for dynamically adjusting the leaning parameters.

Quaternionic and complex-valued Support Vector Regression for Equalization and Function Approximation

Support vector regressors (SVRs) are a class of nonlinear regressor inspired by Vapniks support vector (SV) method for pattern classification. The standard SVR has been successfully applied to real number regression problems such as financial prediction and weather forecasting. However in some applications the domain of the function to be estimated may be more naturally and efficiently expressed using complex numbers (eg. communications channels) or quaternions (eg. 3-dimensional geometrical problems). Since SVRs have previously been proven to be efficient and accurate regressors, the extension of this method to complex numbers and quaternions is of great interest. In the present paper the standard SVR method is extended to cover regression in complex numbers and quaternions. Our method differs from existing approaches in-so-far as the cost function applied in the output space is rotationally invariant, which is important as in most cases it is the magnitude of the error in the output which is important, not the angle. We demonstrate the practical usefulness of this new formulation by considering the problem of communications channel equalization.

Prediction of cystine connectivity using SVM.

One of the major contributors to protein structures is the formation of disulphide bonds between selected pairs of cysteines at oxidized state. Prediction of such disulphide bridges from sequence is challenging given that the possible combination of cysteine pairs as the number of cysteines increases in a protein. Here, we describe a SVM (support vector machine) model for the prediction of cystine connectivity in a protein sequence with and without a priori knowledge on their bonding state. We make use of a new encoding scheme based on physico-chemical properties and statistical features (probability of occurrence of each amino acid residue in different secondary structure states along with PSI-blast profiles). We evaluate our method in SPX (an extended dataset of SP39 (swiss-prot 39) and SP41 (swiss-prot 41) with known disulphide information from PDB) dataset and compare our results with the recursive neural network model described for the same dataset.

A machine learning approach to k-step look-ahead prediction of gait variables from acceleration data

This paper investigates the use of machine learning to predict a sensitive gait parameter based on acceleration information from previous gait cycles. We investigate a k-step look-ahead prediction which attempts to predict gait variable values based on acceleration information in the current gait cycle. The variable is the minimum toe clearance which has been demonstrated to be a sensitive falls risk predictor. Toe clearance data was collected under normal walking conditions and 9 features consisting of peak acceleration and their normalized occurrences times were extracted. A standard least squares estimator, a generalized regression neural network (GRNN) and a support vector regressor (SVR) were trained using 60% of the data to estimate the minimum toe clearance and the remaining 40% was used to validate the model. It was found that when the training data contained data from all subjects (inter-subject) the best GRNN model provided a root mean square error (RMSE) of 2.8mm, the best SVR had RMSE of 2.7mm while the standard least squares linear regression method obtained 3.3mm. When the training and test data consisted of different subject examples (inter-subject) data, the linear SVR demonstrated superior generalization capability (RMSE=3.3mm) compared to other competing models. Validation accuracies up to 5-step look-ahead predictions revealed robust performances for both GRNN and SVR models with no clear degradation in prediction accuracy.