Robert Keight

Machine learning approaches to the application of disease modifying therapy for sickle cell using classification models

This paper discusses the use of machine learning techniques for the classification of medical data, specifically for guiding disease modifying therapies for Sickle Cell. Extensive research has indicated that machine learning approaches generate significant improvements when used for the pre-processing of medical time-series data signals and have assisted in obtaining high accuracy in the classification of medical data. The aim of this paper is to present findings for several classes of learning algorithm for medically related problems. The initial case study addressed in this paper involves classifying the dosage of medication required for the treatment of patients with Sickle Cell Disease. We use different machine learning architectures in order to investigate the accuracy and performance within the case study. The main purpose of applying classification approach is to enable healthcare organisations to provide accurate amount of medication. The results obtained from a range of models during our experiments have shown that of the proposed models, recurrent networks produced inferior results in comparison to conventional feedforward neural networks and the Random Forest model. For our dataset, it was found that the Random Forest Classifier produced the highest levels of performance overall.

Training Neural Networks as Experimental Models: Classifying Biomedical Datasets for Sickle Cell Disease

This paper discusses the use of various type of neural network architectures for the classification of medical data. Extensive research has indicated that neural networks generate significant improvements when used for the pre-processing of medical time-series data signals and have assisted in obtaining high accuracy in the classification of medical data. Up to date, most of hospitals and healthcare sectors in the United Kingdom are using manual approach for analysing patient input for sickle cell disease, which depends on clinician’s experience that can lead to time consuming and stress to patients. The results obtained from a range of models during our experiments have shown that the proposed Back-propagation trained feed-forward neural network classifier generated significantly better outcomes over the other range of classifiers. Using the Receiver Operating Characteristic curve, experiments results showed the following outcomes for our models, in order of best to worst: Back-propagation trained feed-forward neural net classifier: 0.989, Functional Link Neural Network: 0.972, in comparison to the Radial basis neural Network Classifiers with areas of 0.875, and the Voted Perception classifier: 0.766. A Linear Neural Network was used as baseline classifier to illustrate the importance of the previous models, producing an area of 0.849, followed by a random guessing model with an area of 0.524.

Machine learning approaches to predict learning outcomes in Massive open online courses

With the rapid advancements in technology, Massive Open Online Courses (MOOCs) have become the most popular form of online educational delivery, largely due to the removal of geographical and financial barriers for participants. A large number of learners globally enrol in such courses. Despite the flexible accessibility, results indicate that the completion rate is quite low. Educational Data Mining and Learning Analytics are emerging fields of research that aim to enhance the delivery of education through the application of various statistical and machine learning approaches. An extensive literature survey indicates that no significant research is available within the area of MOOC data analysis, in particular considering the behavioural patterns of users. In this paper, therefore, two sets of features, based on learner behavioural patterns, were compared in terms of their suitability for predicting the course outcome of learners participating in MOOCs. Our Exploratory Data Analysis demonstrates that there is strong correlation between click stream actions and successful learner outcomes. Various Machine Learning algorithms have been applied to enhance the accuracy of classifier models. Simulation results from our investigation have shown that Random Forest achieved viable performance for our prediction problem, obtaining the highest performance of the models tested. Conversely, Linear Discriminant Analysis achieved the lowest relative performance, though represented only a marginal reduction in performance relative to the Random Forest.

Artificial Intelligence for Detecting Preterm Uterine Activity in Gynecology and Obstetric Care

Preterm birth brings considerable emotional and economic costs to families and society. However, despite extensive research into understanding the risk factors, the prediction of patient mechanisms and improvements to obstetrical practice, the UK National Health Service still annually spends more than £2.95 billion on this issue. Diagnosis of labour in normal pregnancies is important for minimizing unnecessary hospitalisations, interventions and expenses. Moreover, accurate identification of spontaneous preterm labour would also allow clinicians to start necessary treatments early in women with true labour and avert unnecessary treatment and hospitalisation for women who are simply having preterm contractions, but who are not in true labour. In this research, the Electrohysterography signals have been used to detect preterm births, because Electrohysterography signals provide a strong basis for objective prediction and diagnosis of preterm birth. This has been achieved using an open dataset, which contains 262 records for women who delivered at term and 38 who delivered prematurely. Three different machine learning algorithm were used to identify these records. The results illustrate that the Random Forest performed the best of sensitivity 97%, specificity of 85%, Area under the Receiver Operator curve (AUROC) of 94% and mean square error rate of 14%.

The utilisation of composite machine learning models for the classification of medical datasets for sickle cell disease

The increase growth of health information systems has provided a significant way to deliver great change in medical domains. Up to this date, the majority of medical centres and hospitals continue to use manual approaches for determining the correct medication dosage for sickle cell disease. Such methods depend completely on the experience of medical consultants to determine accurate medication dosages, which can be slow to analyse, time consuming and stressful. The aim of this paper is to provide a robust approach to various applications of machine learning in medical domain problems. The initial case study addressed in this paper considers the classification of medication dosage levels for the treatment of sickle cell disease. This study base on different architectures of machine learning in order to maximise accuracy and performance. The leading motivation for such automated dosage analysis is to enable healthcare organisations to provide accurate therapy recommendations based on previous data. The results obtained from a range of models during our experiments have shown that a composite model, comprising a Neural Network learner, trained using the Levenberg-Marquardt algorithm, combined with a Random Forest learner, produced the best results when compared to other models with an Area under the Curve of 0.995.

Big Data and Data Science Applications for Independent and Healthy Living

Paralleling the state of human progress, developments in healthcare reflect a deeply entrained drive to improve the parameters governing our own existence, including both those which threaten to disrupt our biological functions and followed by those which limit our ability to improve the effectiveness of the former [1, 2]. The technology of the past has allowed us to improve the conditions of our environment and to undertake limited medical interventions in the absence of a direct understanding of disease-causing mechanisms [2]. It is the arrival of the modern era that has opened unprecedented understanding of biological systems and disease mechanisms [3–9], yet such depth of knowledge has also brought a wider realisation of the full complexity and scale of the systems responsible for the biological processes underpinning our existence [10–13]. It is clear that in order to rise to the unprecedented challenges presented by such novel domains, the methods at our disposal must be advanced accordingly to support the changing nature of our task frameworks. The idea that representable forms of information processing may underpin familiar (and novel) forms of intelligence, such as the human brain, raises the possibility that intelligence itself may be practically simulated in alternative settings, for example, via computation, providing a capacity to sustainably address problems of arbitrary complexity. The field of intelligent systems, a research direction within the wider field of artificial intelligence (AI), is concerned with enabling the computational resources of today for the construction of systems that may respond to problems through intelligent abstractions, whose parameters differ from human cognition. Through the combination of computational infrastructure and the patterns of intelligence in this way, it is conceivable that the frontiers of healthcare and medicine may be sustainably advanced to address the broad challenges that underpin the current era.

Towards the discrimination of primary and secondary headache: An intelligent systems approach

We consider the use of intelligent systems to address the long-standing medical problem of diagnostic differentiation between harmful (secondary) and benign (primary) headache conditions. In secondary headaches, conditions are caused by an underlying pathology, in contrast to primary headaches where the production of pain represents the sole constituent of the disorder. Conventional diagnostic paradigms carry an unacceptable risk of misdiagnosis, leaving patients open to potentially catastrophic consequences. Intelligent systems approaches, grounded in artificial intelligence, are adopted in this study as a potential means to unite contributions from multiple settings, including medicine, the life sciences, pervasive computation, sensor technologies, and autonomous intelligent agency, in the fight against headache uncertainty. In this paper, we therefore present the first steps in our research towards a data intensive, unified approach to headache dichotomisation. We begin by presenting a background to headache and its classification, followed by analysis of the space of confounding symptoms, in addition to the problem of primary and secondary condition discrimination. Finally, we proceed to report results of a preliminary case study, in which the epileptic seizure is considered as a manifestation of a headache confounding neuropathology. It was found that our classification approach, based on supervised machine learning, represents a promising direction, with a best area under curve test outcome of 0.915. We conclude that intelligent systems, in conjunction with biosignals, could be suitable for classification of a more general set of pathologies, while facilitating the medicalisation of arbitrary settings.

A Performance Evaluation of Systematic Analysis for Combining Multi-class Models for Sickle Cell Disorder Data Sets

Machine learning approach is considered as a field of science aiming specifically to extract knowledge from the data sets. The main aim of this study is to provide a sophisticate model to difference applications of machine learning models for medically related problems. We attempt for classifying the amount of medications for each patient with Sickle Cell disorder. We present a new technique to combine two classifiers between the Levenberg-Marquartdt training algorithm and the k-nearest neighbours algorithm. In this paper, we introduce multi-class label classification problem in order to obtain training and testing methods for each models along with other performance evaluations. In machine learning, the models utilise a training sets in association with building a classifier that provide a reliable classification. This research discusses different aspects of machine learning approaches for the classification of biomedical data. We are mainly focus on the multi-class label classification problem where many number of classes are available in the data sets. Results have indicated that for the machine learning models tested, the combination classifiers were found to yield considerably better results over the range of performance measures that been selected for this research.

An Intelligent Systems Approach to Primary Headache Diagnosis

In this study, the problem of primary headache diagnosis is considered, referring to multiple frames of reference, including the complexity characteristics of living systems, the limitation of human information processing, the enduring nature of headache throughout history, and the potential for intelligent systems paradigms to both broaden and deepen the scope of such diagnostic solutions. In particular, the use of machine learning is recruited for this study, for which a dataset of 836 primary headache cases is evaluated, originating from two medical centres located in Turkey. Five primary headache classes were derived from the data obtained, namely Tension Type Headache (TTH), Chronic Tension Type Headache (CTTH), Migraine with Aura (MwA), Migraine without Aura (MwoA), followed by Trigeminal Autonomic Cephalalgia (TAC). A total of 9 machine learning based classifiers, ranging from linear to non-linear ensembles, in addition to 1 random baseline procedure, were evaluated within a supervised learning setting, yielding highest performance outcomes of AUC 0.985, sensitivity 1, and specificity 0.966. The study concludes that modern computing platforms represent a promising setting through which to realise intelligent solutions, enabling the space of analytical operations needed to drive forward diagnostic capability in the primary headache domain and beyond.

Towards the Differentiation of Initial and Final Retention in Massive Open Online Courses

Following an accelerating pace of technological change, Massive Open Online Courses (MOOCs) have emerged as a popular educational delivery platform, leveraging ubiquitous connectivity and computing power to overcome longstanding geographical and financial barriers to education. Consequently, the demographic reach of education delivery is extended towards a global online audience, facilitating learning and development for a continually expanding portion of the world population. However, an extensive literature review indicates that the low completion rate is a major issue related to MOOCs. This is considered to be a lack of person to person interaction between instructors and learners on such courses and, the ability of tutors to monitor learners is impaired, often leading to learner withdrawals. To address this problem, learner drop out patterns across five courses offered by Harvard and MIT universities are investigated in this paper. Learning Analytics is applied to address key factors behind participant dropout events through the comparison of attrition during the first and last weeks of each course. The results show that the attrition of participants during the first week of the course is higher than during the last week, low percentages of learners’ attrition are found prior to course closing dates. This could indicate that assessment fees may not represent a significant reason for learners’ withdrawal. We introduce supervised machine learning algorithms for the analysis of learner retention and attrition within a MOOC platform. Results show that machine learning represents a viable direction for the predictive analysis of MOOCs outcomes, with the highest performances yielded by Boosted Tree classification for initial attrition and Neural Network based classification for final attrition.