Search

Highlights • Degree of adoption of decision support tools is improved by knowledge integration. • 5 abstraction layers based Expert systems allows local and global reasoning. • Guidelines indicate a specific decision support system for each sleep stage. • Turing-Machine like decision process reflects dynamic of sleep. • Evaluation should be conducted using inter-rater agreement rates. Abstract Automatic sleep staging is challenging since several issues need to be addressed. Traditional approaches from literature do not satisfy medical experts since they do not reflect the cognitive process they perform when scoring polysomnographic curves. We propose a new approach that is based on the implementation of medical knowledge by symbolic fusion. Medical knowledge coming from the international clinical practice guidelines for sleep medicine is formalized as a five-layer framework dedicated to data abstraction in order to deliver local and global propositions and support the interpretation of polysomnographic curves. Firstly, features are extracted from raw curves. Then these features are combined to recognize sleep events in accordance with guidelines. Sleep events are then fused into the criteria required to recognize the different sleep stages. Sleep is not homogeneous through the night. The physiological events observed during the night follow a dynamic that needs to be included into an automatic sleep staging system. In order to take this into account, decision rules are selected and applied to recognize a sleep stage according to the current context. Thereby, transitions are considered with interest. In this paper, we propose to use a Turing Machine-like decision process to handle transitions. To interpret the local observations and properly score a given state, the previous state which has been stored in a specific register is used as a context. One of the advantages of following the principles of symbolic fusion is to benefit from the full traceability of the decision. Hence, it makes possible to discuss each final — or intermediate — decision with an expert and check for relevance. The preliminary results are encouraging since agreement rates provided between decisions taken by our automatic approach and human experts are similar to those measured between human experts (average agreement rate = 54.60% / average Cohen’s kappa = 0.40) on a dataset of 131 full polysomnographic recordings. [ABSTRACT FROM AUTHOR]

Expert systems for diabetes care need to automatically evaluate glycaemia measurements in relationship to meals to correctly determine patients’ metabolic condition and generate recommendations about therapy adjustments. Most glucose meters allow patients to manually label each measurement with a meal tag, but as this utility is not always used, a completion procedure is needed. Classification methods are usually based on predefined mealtimes and present insufficient accuracy that might affect the automatic data analysis. Expert systems in diabetes require a reliable method to manage incomplete glycaemia data so that they can determine if patients’ metabolic condition is altered due to a specific meal or due to an extended fasting period. This paper presents the design and application of a classification module to automatically assign the appropriate meal and ‘moment of measurement’ to incomplete glycaemia data. Different machine learning techniques were studied in order to design the best classification algorithm in terms of accuracy. The selected classifier was implemented with a C4.5 decision tree with 7 input features selected with a wrapper evaluator and the genetic search algorithm, which achieved 95.45% of accuracy with the training set on cross-validation. The classification module was integrated in the Sinedie expert system for gestational diabetes care and was evaluated in a clinical environment for 8 months with 42 patients. A total of 7,113 glycaemia measurements were uploaded by patients into the Sinedie system and were completed by the “classification module”. The 98.79% of the measurements were correctly classified, while patients modified the automatic classification of 1.21% of them. Classification results were improved by 21.04% compared to a classification based on predefined mealtimes. The automatic classification of glycaemia measurements minimizes the patient's intervention, allows structuring measurements in relationship to meals and makes automatic data interpretation by expert systems more reliable. [ABSTRACT FROM AUTHOR]

The rapidly growing world energy use already has concerns over the exhaustion of energy resources and heavy environmental impacts. As a result of these concerns, a trend of green and smart cities has been increasing. To respond to this increasing trend of smart cities with buildings every time more complex, in this paper we have proposed a new method to solve energy inefficiencies detection problem in smart buildings. This solution is based on a rule-based system developed through data mining techniques and applying the knowledge of energy efficiency experts. A set of useful energy efficiency indicators is also proposed to detect anomalies. The data mining system is developed through the knowledge extracted by a full set of building sensors. So, the results of this process provide a set of rules that are used as a part of a decision support system for the optimisation of energy consumption and the detection of anomalies in smart buildings. [ABSTRACT FROM AUTHOR]

Service monitoring in federated clouds generates large scale QoS time series data with various unknown, frequent and abnormal patterns. This could be associated with inaccurate resource provisioning and avoid violations through predictive and preventive actions. A sufficient intelligence in the form of expert system for decision support is needed in such situations. Therefore, the main challenge here is to efficiently discover unknown frequent and abnormal patterns from QoS time series data of federated clouds. On the other hand, QoS time series data in federated clouds is unlabeled and consists of frequent and abnormal structures. Studies showed that clustering is the most common and efficient method to discover interesting patterns and structures from unlabeled data. But, clustering is normally associated with time overhead that should be optimized as well as accuracy issues mainly in connection with convergence and finding an optimum number of clusters. This work proposes a new genetic based clustering algorithm that shows better accuracy and speed in comparison to state-of-the-art methods. Furthermore, the proposed algorithm can find the optimum number of clusters concurrently with the clustering itself. Achieved accuracy and convergence of the proposed method in the experimental results assure its use in expert systems, mainly for resource provisioning and further autonomous decision making situations in federated clouds. In addition to the scientific impact of this paper, the proposed method can be used by federated cloud service providers in practice. • Shape based clustering of large scale QoS time series data in federated clouds. • Genetic based clustering using variable chromosome length. • Crossover and Mutation operation using the meta data of time series. • Simultaneous lookup of the best number of clusters while clustering. • Accurate and faster clustering performance compared to state-of-the-art methods. [ABSTRACT FROM AUTHOR]

• We deal with triage prediction systems using machine learning in pediatric patients. • A very large in-house dataset was used for model construction and validation. • We propose a data re-labeling process aiming to improve clinical outcome prediction. • Random forest provides best results over a bunch of multi-class performance metrics. • This result outperforms traditional Triage models. Triage is a critical process in hospital emergency departments (ED). Specifically, we consider how to achieve fast and accurate patient Triage in the ED of a pediatric hospital. The goal of this paper is to establish methodological best practices for the application of machine learning (ML) to Triage in pediatric ED, providing a comprehensive comparison of the performance of ML techniques over a large dataset. Our work is among the first attempts in this direction. Following very recent works in the literature, we use the clinical outcome of a case as its label for supervised ML model training, instead of the more uncertain labels provided by experts. The experimental dataset contains the records along 3 years of operation of the hospital ED. It consists of 189,718 patients visits to the hospital. The clinical outcome of 9271 cases (4.98%) wa hospital admission, therefore our dataset is highly class imbalanced. Our reported performance comparison results focus on four ML models: Deep Learning (DL), Random Forest (RF), Naive Bayes (NB) and Support Vector Machines (SVM). Data preprocessing includes class imbalance correction, and case re-labeling. We use different well known metrics to evaluate performance of ML models in three different experimental settings: (a) classification of each case into the standard five Triage urgency levels, (b) discrimination of high versus low case severity according to its clinical outcome, and (c) comparison of the number of patients assigned to each standard Triage urgency level against the Triage rule based expert system currently in use at the hospital. RF achieved greater AUC, accuracy, PPV and specificity than the other models in the dychotomic classification experiments. On the implementation side, our study shows that ML predictive models trained according to clinical outcomes, provide better Triage performance than the current rule based expert system in operation at the hospital. [ABSTRACT FROM AUTHOR]