Machine learning approaches to analysing textual injury surveillance data: A systematic review

Objective To synthesise recent research on the use of machine learning approaches to mining textual injury surveillance data. Design Systematic review. Data sources The electronic databases which were searched included PubMed, Cinahl, Medline, Google Scholar, and Proquest. The bibliography of all relevant articles was examined and associated articles were identified using a snowballing technique. Selection criteria For inclusion, articles were required to meet the following criteria: (a) used a health-related database, (b) focused on injury-related cases, AND used machine learning approaches to analyse textual data. Methods The papers identified through the search were screened resulting in 16 papers selected for review. Articles were reviewed to describe the databases and methodology used, the strength and limitations of different techniques, and quality assurance approaches used. Due to heterogeneity between studies meta-analysis was not performed. Results Occupational injuries were the focus of half of the machine learning studies and the most common methods described were Bayesian probability or Bayesian network based methods to either predict injury categories or extract common injury scenarios. Models were evaluated through either comparison with gold standard data or content expert evaluation or statistical measures of quality. Machine learning was found to provide high precision and accuracy when predicting a small number of categories, was valuable for visualisation of injury patterns and prediction of future outcomes. However, difficulties related to generalizability, source data quality, complexity of models and integration of content and technical knowledge were discussed. Conclusions The use of narrative text for injury surveillance has grown in popularity, complexity and quality over recent years. With advances in data mining techniques, increased capacity for analysis of large databases, and involvement of computer scientists in the injury prevention field, along with more comprehensive use and description of quality assurance methods in text mining approaches, it is likely that we will see a continued growth and advancement in knowledge of text mining in the injury field.

Centroid training to achieve effective text classification

Traditional text classification technology based on machine learning and data mining techniques has made a big progress. However, it is still a big problem on how to draw an exact decision boundary between relevant and irrelevant objects in binary classification due to much uncertainty produced in the process of the traditional algorithms. The proposed model CTTC (Centroid Training for Text Classification) aims to build an uncertainty boundary to absorb as many indeterminate objects as possible so as to elevate the certainty of the relevant and irrelevant groups through the centroid clustering and training process. The clustering starts from the two training subsets labelled as relevant or irrelevant respectively to create two principal centroid vectors by which all the training samples are further separated into three groups: POS, NEG and BND, with all the indeterminate objects absorbed into the uncertain decision boundary BND. Two pairs of centroid vectors are proposed to be trained and optimized through the subsequent iterative multi-learning process, all of which are proposed to collaboratively help predict the polarities of the incoming objects thereafter. For the assessment of the proposed model, F1 and Accuracy have been chosen as the key evaluation measures. We stress the F1 measure because it can display the overall performance improvement of the final classifier better than Accuracy. A large number of experiments have been completed using the proposed model on the Reuters Corpus Volume 1 (RCV1) which is important standard dataset in the field. The experiment results show that the proposed model has significantly improved the binary text classification performance in both F1 and Accuracy compared with three other influential baseline models.

Large scale monitoring of crowds and building utilisation: A new database and distributed approach

Public buildings and large infrastructure are typically monitored by tens or hundreds of cameras, all capturing different physical spaces and observing different types of interactions and behaviours. However to date, in large part due to limited data availability, crowd monitoring and operational surveillance research has focused on single camera scenarios which are not representative of real-world applications. In this paper we present a new, publicly available database for large scale crowd surveillance. Footage from 12 cameras for a full work day covering the main floor of a busy university campus building, including an internal and external foyer, elevator foyers, and the main external approach are provided; alongside annotation for crowd counting (single or multi-camera) and pedestrian flow analysis for 10 and 6 sites respectively. We describe how this large dataset can be used to perform distributed monitoring of building utilisation, and demonstrate the potential of this dataset to understand and learn the relationship between different areas of a building.

Real-time traffic state estimation in urban corridors from heterogeneous data

In recent years, rapid advances in information technology have led to various data collection systems which are enriching the sources of empirical data for use in transport systems. Currently, traffic data are collected through various sensors including loop detectors, probe vehicles, cell-phones, Bluetooth, video cameras, remote sensing and public transport smart cards. It has been argued that combining the complementary information from multiple sources will generally result in better accuracy, increased robustness and reduced ambiguity. Despite the fact that there have been substantial advances in data assimilation techniques to reconstruct and predict the traffic state from multiple data sources, such methods are generally data-driven and do not fully utilize the power of traffic models. Furthermore, the existing methods are still limited to freeway networks and are not yet applicable in the urban context due to the enhanced complexity of the flow behavior. The main traffic phenomena on urban links are generally caused by the boundary conditions at intersections, un-signalized or signalized, at which the switching of the traffic lights and the turning maneuvers of the road users lead to shock-wave phenomena that propagate upstream of the intersections. This paper develops a new model-based methodology to build up a real-time traffic prediction model for arterial corridors using data from multiple sources, particularly from loop detectors and partial observations from Bluetooth and GPS devices.

Aspect-based opinion mining from product reviews using conditional random fields

Product reviews are the foremost source of information for customers and manufacturers to help them make appropriate purchasing and production decisions. Natural language data is typically very sparse; the most common words are those that do not carry a lot of semantic content, and occurrences of any particular content-bearing word are rare, while co-occurrences of these words are rarer. Mining product aspects, along with corresponding opinions, is essential for Aspect-Based Opinion Mining (ABOM) as a result of the e-commerce revolution. Therefore, the need for automatic mining of reviews has reached a peak. In this work, we deal with ABOM as sequence labelling problem and propose a supervised extraction method to identify product aspects and corresponding opinions. We use Conditional Random Fields (CRFs) to solve the extraction problem and propose a feature function to enhance accuracy. The proposed method is evaluated using two different datasets. We also evaluate the effectiveness of feature function and the optimisation through multiple experiments.

Computer-assisted sampling of acoustic data for more efficient determination of bird species richness

Bird species richness survey is one of the most intriguing ecological topics for evaluating environmental health. Here, bird species richness denotes the number of unique bird species in a particular area. Factors affecting the investigation of bird species richness include weather, observation bias, and most importantly, the prohibitive costs of conducting surveys at large spatiotemporal scales. Thanks to advances in recording techniques, these problems have been alleviated by deploying sensors for acoustic data collection. Although automated detection techniques have been introduced to identify various bird species, the innate complexity of bird vocalizations, the background noise present in the recording and the escalating volumes of acoustic data pose a challenging task on determination of bird species richness. In this paper we proposed a two-step computer-assisted sampling approach for determining bird species richness in one-day acoustic data. First, a classification model is built based on acoustic indices for filtering out minutes that contain few bird species. Then the classified bird minutes are ordered by an acoustic index and the redundant temporal minutes are removed from the ranked minute sequence. The experimental results show that our method is more efficient in directing experts for determination of bird species compared with the previous methods.

Enhancement of relevant features for text mining

With the explosion of information resources, there is an imminent need to understand interesting text features or topics in massive text information. This thesis proposes a theoretical model to accurately weight specific text features, such as patterns and n-grams. The proposed model achieves impressive performance in two data collections, Reuters Corpus Volume 1 (RCV1) and Reuters 21578.

GRDB: A general purpose relational database system

This article discusses the design and development of GRDB (General Purpose Relational Data Base System) which has been implemented on a DEC-1090 system in Pascal. GRDB is a general purpose database system designed to be completely independent of the nature of data to be handled, since it is not tailored to the specific requirements of any particular enterprise. It can handle different types of data such as variable length records and textual data. Apart from the usual database facilities such as data definition and data manipulation, GRDB supports User Definition Language (UDL) and Security definition language. These facilities are provided through a SEQUEL-like General Purpose Query Language (GQL). GRDB provides adequate protection facilities up to the relation level. The concept of “security matrix” has been made use of to provide database protection. The concept of Unique IDentification number (UID) and Password is made use of to ensure user identification and authentication. The concept of static integrity constraints has been used to ensure data integrity. Considerable efforts have been made to improve the response time through indexing on the data files and query optimisation. GRDB is designed for an interactive use but alternate provision has been made for its use through batch mode also. A typical Air Force application (consisting of data about personnel, inventory control, and maintenance planning) has been used to test GRDB and it has been found to perform satisfactorily.

The navigation and visualisation of environmental audio using zooming spectrograms

Acoustic recordings play an increasingly important role in monitoring terrestrial and aquatic environments. However, rapid advances in technology make it possible to accumulate thousands of hours of recordings, more than ecologists can ever listen to. Our approach to this big-data challenge is to visualize the content of long-duration audio recordings on multiple scales, from minutes, hours, days to years. The visualization should facilitate navigation and yield ecologically meaningful information prior to listening to the audio. To construct images, we calculate acoustic indices, statistics that describe the distribution of acoustic energy and reflect content of ecological interest. We combine various indices to produce false-color spectrogram images that reveal acoustic content and facilitate navigation. The technical challenge we investigate in this work is how to navigate recordings that are days or even months in duration. We introduce a method of zooming through multiple temporal scales, analogous to Google Maps. However, the “landscape” to be navigated is not geographical and not therefore intrinsically visual, but rather a graphical representation of the underlying audio. We describe solutions to navigating spectrograms that range over three orders of magnitude of temporal scale. We make three sets of observations: 1. We determine that at least ten intermediate scale steps are required to zoom over three orders of magnitude of temporal scale; 2. We determine that three different visual representations are required to cover the range of temporal scales; 3. We present a solution to the problem of maintaining visual continuity when stepping between different visual representations. Finally, we demonstrate the utility of the approach with four case studies.

A flexible hierarchical approach for facial age estimation based on multiple features

Age estimation from facial images is increasingly receiving attention to solve age-based access control, age-adaptive targeted marketing, amongst other applications. Since even humans can be induced in error due to the complex biological processes involved, finding a robust method remains a research challenge today. In this paper, we propose a new framework for the integration of Active Appearance Models (AAM), Local Binary Patterns (LBP), Gabor wavelets (GW) and Local Phase Quantization (LPQ) in order to obtain a highly discriminative feature representation which is able to model shape, appearance, wrinkles and skin spots. In addition, this paper proposes a novel flexible hierarchical age estimation approach consisting of a multi-class Support Vector Machine (SVM) to classify a subject into an age group followed by a Support Vector Regression (SVR) to estimate a specific age. The errors that may happen in the classification step, caused by the hard boundaries between age classes, are compensated in the specific age estimation by a flexible overlapping of the age ranges. The performance of the proposed approach was evaluated on FG-NET Aging and MORPH Album 2 datasets and a mean absolute error (MAE) of 4.50 and 5.86 years was achieved respectively. The robustness of the proposed approach was also evaluated on a merge of both datasets and a MAE of 5.20 years was achieved. Furthermore, we have also compared the age estimation made by humans with the proposed approach and it has shown that the machine outperforms humans. The proposed approach is competitive with current state-of-the-art and it provides an additional robustness to blur, lighting and expression variance brought about by the local phase features.

Data distribution and task scheduling for distributed computing of all-to-all comparison problems

This research studied distributed computing of all-to-all comparison problems with big data sets. The thesis formalised the problem, and developed a high-performance and scalable computing framework with a programming model, data distribution strategies and task scheduling policies to solve the problem. The study considered storage usage, data locality and load balancing for performance improvement in solving the problem. The research outcomes can be applied in bioinformatics, biometrics and data mining and other domains in which all-to-all comparisons are a typical computing pattern.

Computationally Efficient Methods for MDL-Optimal Density Estimation and Data Clustering

The Minimum Description Length (MDL) principle is a general, well-founded theoretical formalization of statistical modeling. The most important notion of MDL is the stochastic complexity, which can be interpreted as the shortest description length of a given sample of data relative to a model class. The exact definition of the stochastic complexity has gone through several evolutionary steps. The latest instantation is based on the so-called Normalized Maximum Likelihood (NML) distribution which has been shown to possess several important theoretical properties. However, the applications of this modern version of the MDL have been quite rare because of computational complexity problems, i.e., for discrete data, the definition of NML involves an exponential sum, and in the case of continuous data, a multi-dimensional integral usually infeasible to evaluate or even approximate accurately. In this doctoral dissertation, we present mathematical techniques for computing NML efficiently for some model families involving discrete data. We also show how these techniques can be used to apply MDL in two practical applications: histogram density estimation and clustering of multi-dimensional data.

Using a glow graph to represent data flow and dependency in event logs

The idea of extracting knowledge in process mining is a descendant of data mining. Both mining disciplines emphasise data flow and relations among elements in the data. Unfortunately, challenges have been encountered when working with the data flow and relations. One of the challenges is that the representation of the data flow between a pair of elements or tasks is insufficiently simplified and formulated, as it considers only a one-to-one data flow relation. In this paper, we discuss how the effectiveness of knowledge representation can be extended in both disciplines. To this end, we introduce a new representation of the data flow and dependency formulation using a flow graph. The flow graph solves the issue of the insufficiency of presenting other relation types, such as many-to-one and one-to-many relations. As an experiment, a new evaluation framework is applied to the Teleclaim process in order to show how this method can provide us with more precise results when compared with other representations.

Multi-GNSS precise point positioning with raw single-frequency and dual-frequency measurement models

The emergence of multiple satellite navigation systems, including BDS, Galileo, modernized GPS, and GLONASS, brings great opportunities and challenges for precise point positioning (PPP). We study the contributions of various GNSS combinations to PPP performance based on undifferenced or raw observations, in which the signal delays and ionospheric delays must be considered. A priori ionospheric knowledge, such as regional or global corrections, strengthens the estimation of ionospheric delay parameters. The undifferenced models are generally more suitable for single-, dual-, or multi-frequency data processing for single or combined GNSS constellations. Another advantage over ionospheric-free PPP models is that undifferenced models avoid noise amplification by linear combinations. Extensive performance evaluations are conducted with multi-GNSS data sets collected from 105 MGEX stations in July 2014. Dual-frequency PPP results from each single constellation show that the convergence time of undifferenced PPP solution is usually shorter than that of ionospheric-free PPP solutions, while the positioning accuracy of undifferenced PPP shows more improvement for the GLONASS system. In addition, the GLONASS undifferenced PPP results demonstrate performance advantages in high latitude areas, while this impact is less obvious in the GPS/GLONASS combined configuration. The results have also indicated that the BDS GEO satellites have negative impacts on the undifferenced PPP performance given the current “poor” orbit and clock knowledge of GEO satellites. More generally, the multi-GNSS undifferenced PPP results have shown improvements in the convergence time by more than 60 % in both the single- and dual-frequency PPP results, while the positioning accuracy after convergence indicates no significant improvements for the dual-frequency PPP solutions, but an improvement of about 25 % on average for the single-frequency PPP solutions.

Selection of intrusion detection system threshold bounds for effective sensor fusion

The motivation behind the fusion of Intrusion Detection Systems was the realization that with the increasing traffic and increasing complexity of attacks, none of the present day stand-alone Intrusion Detection Systems can meet the high demand for a very high detection rate and an extremely low false positive rate. Multi-sensor fusion can be used to meet these requirements by a refinement of the combined response of different Intrusion Detection Systems. In this paper, we show the design technique of sensor fusion to best utilize the useful response from multiple sensors by an appropriate adjustment of the fusion threshold. The threshold is generally chosen according to the past experiences or by an expert system. In this paper, we show that the choice of the threshold bounds according to the Chebyshev inequality principle performs better. This approach also helps to solve the problem of scalability and has the advantage of failsafe capability. This paper theoretically models the fusion of Intrusion Detection Systems for the purpose of proving the improvement in performance, supplemented with the empirical evaluation. The combination of complementary sensors is shown to detect more attacks than the individual components. Since the individual sensors chosen detect sufficiently different attacks, their result can be merged for improved performance. The combination is done in different ways like (i) taking all the alarms from each system and avoiding duplications, (ii) taking alarms from each system by fixing threshold bounds, and (iii) rule-based fusion with a priori knowledge of the individual sensor performance. A number of evaluation metrics are used, and the results indicate that there is an overall enhancement in the performance of the combined detector using sensor fusion incorporating the threshold bounds and significantly better performance using simple rule-based fusion.