Andrea Kulakov

Application of wavelet neural-networks in wireless sensor networks

Some of the algorithms developed within the artificial neural-networks tradition can be easily adopted to wireless sensor network platforms and in the same time they can meet the requirements for sensor networks like: simple parallel distributed computation, distributed storage, data robustness and auto-classification of sensor readings. Dimensionality reduction, obtained simply from the outputs of the neural-networks clustering algorithms, leads to lower communication costs and energy savings. Two different data aggregation architectures are presented, with algorithms which use wavelets for initial data-processing of the sensory inputs and artificial neural-networks which use unsupervised learning methods for categorization of the sensory inputs. They are analyzed on a data obtained from a set of several motes, equipped with several sensors each. Results from deliberately simulated malfunctioning sensors show the data robustness of these architectures.

Improving Activity Recognition Accuracy in Ambient-Assisted Living Systems by Automated Feature Engineering

Ambient-assisted living (AAL) is promising to become a supplement of the current care models, providing enhanced living experience to people within context-aware homes and smart environments. Activity recognition based on sensory data in AAL systems is an important task because 1) it can be used for estimation of levels of physical activity, 2) it can lead to detecting changes of daily patterns that may indicate an emerging medical condition, or 3) it can be used for detection of accidents and emergencies. To be accepted, AAL systems must be affordable while providing reliable performance. These two factors hugely depend on optimizing the number of utilized sensors and extracting robust features from them. This paper proposes a generic feature engineering method for selecting robust features from a variety of sensors, which can be used for generating reliable classification models. From the originally recorded time series and some newly generated time series [i.e., magnitudes, first derivatives, delta series, and fast Fourier transformation (FFT)-based series], a variety of time and frequency domain features are extracted. Then, using two-phase feature selection, the number of generated features is greatly reduced. Finally, different classification models are trained and evaluated on an independent test set. The proposed method was evaluated on five publicly available data sets, and on all of them, it yielded better accuracy than when using hand-tailored features. The benefits of the proposed systematic feature engineering method are quickly discovering good feature sets for any given task than manually finding ones suitable for a particular task, selecting a small feature set that outperforms manually determined features in both execution time and accuracy, and identification of relevant sensor types and body locations automatically. Ultimately, the proposed method could reduce the cost of AAL systems by facilitating execution of algorithms on devices with limited resources and by using as few sensors as possible.

Architecture for Wireless Sensor and Actor Networks Control and Data Acquisition

Wireless Sensor and Actor Networks (WSANs) have received increased attention from the research community in the recent years. This is mainly because as an extension to Wireless Sensor Networks(WSN), they have the ability to actively participate in the environment trough the actors. However, this introduces new challenges as to how to transfer commands between nodes, actors and central station who may be from different manufacturers and use different communication protocols. Another important aspect is the ability of the WSAN to present the data to the interested party or to receive the command from the operator, and do this with in the simplest and most user friendly way as possible. In this paper we propose architecture for interconnection between different layers of WSANs and the central stations that would allow building a simple interface that would ease the operation with WSANs in view of Control and Data Acquisition.

Robust histogram-based feature engineering of time series data

Collecting data at regular time nowadays is ubiquitous. The most widely used type of data that is being collected and analyzed is financial data and sensor readings. Various businesses have realized that financial time series analysis is a powerful analytical tool that can lead to competitive advantages. Likewise, sensor networks generate time series and if they are properly analyzed can give a better understanding of the processes that are being monitored. In this paper we propose a novel generic histogram-based method for feature engineering of time series data. The preprocessing phase consists of several steps: deseansonalyzing the time series data, modeling the speed of change with first derivatives, and finally calculating histograms. By doing all of those steps the goal is three-fold: achieve invariance to different factors, good modeling of the data and preform significant feature reduction. This method was applied to the AAIA Data Mining Competition 2015, which was concerned with recognition of activities carried out by firefighters by analyzing body sensor network readings. By doing that we were able to score the third place with predictive accuracy of about 83%, which was about 1% worse than the winning solution.

Cluster-based MDS algorithm for nodes localization in wireless sensor networks with irregular topologies

Nodes localization in Wireless Sensor Networks (WSN) has arisen as a very challenging problem in the research community. Most of the applications for WSN are not useful without a priori known nodes positions. One solution to the problem is by adding GPS receivers to each node. Since this is an expensive approach and inapplicable for indoor environments, we need to find an alternative intelligent mechanism for determining nodes location. In this paper, we propose our cluster-based approach of multidimensional scaling (MDS) technique. Our initial experiments show that our algorithm outperforms MDS-MAP[8], particularly for irregular topologies in terms of accuracy.

Intelligent Wireless Sensor Networks Using FuzzyART Neural-Networks

An adaptation of one popular model of neural-networks algorithm (ART model) in the field of wireless sensor networks is demonstrated in this paper. The important advantages of the ART class algorithms such as simple parallel distributed computation, distributed storage, data robustness and auto-classification of sensor readings are confirmed within the proposed architecture consisting of one clusterhead which collects only classified input data from the other units. This architecture provides a high dimensionality reduction and additional communication savings, since only identification numbers of the classified input data are passed to the clusterhead instead of the whole input samples. We have adapted and implemented the FuzzyART neural-network algorithm and used it for initial clustering of the sensor data as a sort of pattern recognition. This adaptation was made specifically for MicaZ sensor motes by solving mainly problems concerning the small memory capacity of the motes. At the final clusterhead -server, the data are stored in a database and the results of the data processing are continuously presented in a classification graph.

Single exponential smoothing method and neural network in one method for time series prediction

The purpose of this paper is to present a new method that combines statistical techniques and neural networks in one method for the better time series prediction. In this paper we presented single exponential smoothing method (statistical technique) merged with feed forward back propagation neural network in one method named as smart single exponential smoothing method (SSESM). The basic idea of the new method is to learn from the mistakes. More specifically, our neural network learns from the mistakes made by the statistical techniques. The mistakes are made by the smoothing parameter, which is constant. In our method, the smoothing parameter is a variable. It is changed according to the prediction of the neural network. Experimental results show that the prediction with a variable smoothing parameter is better than with a constant smoothing parameter

SVM Parameter Tuning with Grid Search and Its Impact on Reduction of Model Over-fitting

In this paper we describe our submission to the IJCRS’15 Data Mining Competition, which is concerned with prediction of dangerous concentrations of methane in longwalls of a Polish coalmine. We address the challenge of building robust classification models with support vector machines (SVMs) that are built from time series data. Moreover, we investigate the impact of parameter tuning of SVMs with grid search on the classification performance and its effect on preventing over-fitting. Our results show improvements of predictive performance with proper parameter tuning but also improved stability of the classification models even when the test data comes from a different time period and class distribution. By applying the proposed method we were able to build a classification model that predicts unseen test data even better than the training data, thus highlighting the non-over-fitting properties of the model. The submitted solution was about 2 % behind the winning solution.

Inductive Logic Programming ILP and Reasoning by Analogy in Context of Embodied Robot Learning

The ability to reason by analogy is essential for many cognitive processes from low-level and high-level perception to categorization. Intuitively, the idea is to use what is already known to explain new observations that appear similar to old knowledge. In a sense, it is opposite of induction, where to explain the observations one comes up with new hypotheses/theories. Therefore, a system capable of both types of reasoning would be superior. In this paper, the authors present an overview of Inductive Logic Programming ILP systems that use reasoning by analogy and discuss the results of combining Analogical Prediction with an ILP system, showing that, for some cases, it is possible to improve significantly the learning speed of the ILP system. This paper will examine the problems that arise in the context of a physically embodied robot that tries to learn regularities in its environment.

Transformation of nominal features into numeric in supervised multi-class problems based on the weight of evidence parameter

Machine learning has received increased interest by both the scientific community and the industry. Most of the machine learning algorithms rely on certain distance metrics that can only be applied to numeric data. This becomes a problem in complex datasets that contain heterogeneous data consisted of numeric and nominal (i.e. categorical) features. Thus the need of transformation from nominal to numeric data. Weight of evidence (WoE) is one of the parameters that can be used for transformation of the nominal features to numeric. In this paper we describe a method that uses WoE to transform the features. Although the applicability of this method is researched to some extent, in this paper we extend its applicability for multi-class problems, which is a novelty. We compared it with the method that generates dummy features. We test both methods on binary and multi-class classification problems with different machine learning algorithms. Our experiments show that the WoE based transformation generates smaller number of features compared to the technique based on generation of dummy features while also improving the classification accuracy, reducing memory complexity and shortening the execution time. Be that as it may, we also point out some of its weaknesses and make some recommendations when to use the method based on dummy features generation instead.