Falls are a leading cause of injury and accidental injury death worldwide. Fall-risk prevention techniques exist but fall-risk identification can be difficult. While clinical assessment tools are the standard for identifying fall risk, wearable-sensors and machine learning could improve outcomes with automated and efficient techniques. Machine learning research has focused on older adults. Since people with lower limb amputations have greater falling and injury risk than the elderly, research is needed to evaluate these approaches with the amputee population. In this thesis, random forest and fully connected feedforward artificial neural network (ANN) machine learning models were developed and optimized for fall-risk identification in amputee populations, using smartphone sensor data (phone at posterior pelvis) from 89 people with various levels of lower-limb amputation who completed a 6-minute walk test (6MWT). The best model was a random forest with 500 trees, using turn data and a feature set selected using correlation-based feature selection (81.3% accuracy, 57.2% sensitivity, 94.9% specificity, 0.59 Matthews correlation coefficient, 0.83 F1 score). After extensive ANN optimization with the best ranked 50 features from an Extra Trees Classifier, the best ANN model achieved 69.7% accuracy, 53.1% sensitivity, 78.9% specificity, 0.33 Matthews correlation coefficient, and 0.62 F1 score. Features from a single smartphone during a 6MWT can be used with random forest machine learning for fall-risk classification in lower limb amputees. Model performance was similarly effective or better than the Timed Up and Go and Four Square Step Test. This model could be used clinically to identify fall-risk individuals during a 6MWT, thereby finding people who were not intended for fall screening. Since model specificity was very high, the risk of accidentally misclassifying people who are a no fall-risk individual is quite low, and few people would incorrectly be entered into fall mitigation programs based on the test outcomes.

Artificial neural networks are demanding to computational power of a computer. Increasing their learning speed could mean new posibilities for research or aplication of the algorithm. And that is a purpose of this thesis. The usage of graphics processing units for neural networks learning is one way how to achieve above mentioned goals. This thesis is offering a survey of theoretical background and consequently implementation of a software library for neural networks learning with a Backpropagation algorithm with a support of acceleration on graphics processing unit.

Without improved water management the global population could be facing serious water shortages. River flow discharge rates are one factor that could contribute to improving water management, being able to predict a forecasted river flow value would provide support in the management of water resources. This research investigates the use of an artificial neural network (ANN) to create a model that predicts river flow gauge values. The Driel Barrage monitoring station on the Thukela river in South Africa was used as a case study. The research makes use of data from the Department of Water and Sanitation (DWS) and weather forecast data from the European Center For Medium- Range Forecasts (ECMWF) to train the predictive model. An evaluation of the ANN model identified that the model is highly sensitive to selected weather parameters and is sensitive to the initial weights used in the ANN. These were overcome using an ANN ensemble and selective scenarios to identify the best weather parameters to use as input into the ANN model. Five weather parameters and a correlation coefficient cut-off value produced the most accurate prediction by the ANN. The research found that ANNs can be used for predicting river flow gauge values but to improve the results a greater ensemble, additional data and different ANN structures may create a better performing model. For the ANN model to be used in practice the research needs to be extended to evaluate the whole catchment area and a range of rivers in South Africa. / Dissertation (MSc)--University of Pretoria, 2017. / Geography, Geoinformatics and Meteorology / MSc / Unrestricted

As the impacts of climate change intensify, businesses are increasingly committing to ambitious sustainable development goals, yet an enduring disconnect remains between corporate sustainability activities and declining global environment and society. This study adopts a complexity view that reductionism associated with Newtonian thinking has played a key role in creating many of the sustainability issues now faced by humanity. This dissertation departs from the premise that sustainability needs to be integrated into an organisation and uses a complexity view to argue that corporate sustainability is a co-evolutionary process of emergence. Whilst many studies have examined how sustainability can be integrated into a business, less is known about corporate sustainability as an emergent process. To address the knowledge gap, this research answered three questions: (1) How does sustainability emerge in financial institutions? (2) What is the role of coherence in the emergence of sustainability? and (3) What conditions enable the emergence of sustainability? A mixed method sequential design was used. In the initial quantitative strand of the research, a holistic business assessment survey based on integral theory was implemented in two financial services organisations in Southern Africa. The results were analysed using self-organising maps and explored in narrative interviews in the subsequent qualitative strand of the research. The study makes three contributions to our understanding of emergence in corporate sustainability. First, by proposing four modes by which corporate sustainability is enacted; these elucidate how integral domains are enacted in corporate sustainability. Second, by clarifying the process of emergence by articulating how zones of coherence emerge between embodied and embedded dimensions. Third, by explaining how the shift to corporate sustainability occurs by means of four conditions. These contributions serve to advance our understanding of corporate sustainability as a fundamental shift in the functioning of an organisation towards coevolutionary self-organisation. It is recommended that corporate sustainability is holistically cultivated to support emergence and self-organisation, rather than being integrated through a linear process of change.

No description available.

Prediction of stock prices is an important financial problem that is receiving increased attention in the field of artificial intelligence. Many different neural network and hybrid models for obtaining accurate prediction results have been proposed during the last few years in an attempt to outperform the traditional linear and nonlinear approaches. This study evaluates the performance of three different hybrid neural network models used for one-day stock close price prediction; a pre-processed evolutionary Levenberg-Marquardt neural network, Bayesian regularized artificial neural network and neural network with technical- and fractal analysis. It was also determined which of the three outperformed the others. The performance evaluation and comparison of the models are done using statistical error measures for accuracy; mean square error, symmetric mean absolute percentage error and point of change in direction. The results indicate good performance values for the Bayesian regularized artificial neural network, and varied performance for the others. Using the Friedman test, one model clearly is different in its performance relative to the others, probably the above mentioned model. The results for two of the models showed a large standard deviation of the error measurements which indicates that the results are not entirely reliable.

El texto completo de este trabajo no está disponible en el Repositorio Académico UPC por restricciones de la casa editorial donde ha sido publicado. / The aim of this article is to use artificial neural networks (ANN) to perform the structural design of confined masonry buildings. ANN is easy to operate and allows to reduce the time and cost of seismic designs. To generate the artificial neural network, training models (traditional confined masonry designs) are used to identify the input and output parameters. From this, the final architecture and activation functions are defined for each layer of the ANN. Finally, ANN training is carried out using the backpropagation algorithm to obtain the matrix of weights and thresholds that allow the network to operate and provide preliminary structural designs with a 10% margin of error, with respect to the traditional design, in the dimensions and reinforcements of the structural elements.

Internal combustion engines, used for light duty transportation, represent a major role in mobility, contributing 28.6% to CO$_2$ emissions worldwide. To mitigate environmental impact and ease the transition to clean technologies, the search for more efficient, less polluting engines has been demanded, and unique tools are necessary to meet the constantly upgraded policies. Hence, data-driven approaches that emulate current vehicles represent a valuable contribution to the improvement of engine performance. Dynamometer tests of commercial engines are open-data, and a dependable source for understanding on-road behavior of several vehicle variables. Artificial neural network (ANN) algorithms, a subset of machine learning, have received considerable attention recently given their wide number of applications and the possibility to provide accurate data-driven approximations. This work describes a methodology for applying ANN’s to predict emissions, efficiency, and fuel consumption in combustion engines using dynamometer test data, and to extrapolate its use in new technologies. The procedure is also applied to a hybrid vehicle case study. The proposed methodology accurately generates ANN’s for the prediction of brake thermal efficiency (BTE), brake specific fuel consumption (BSFC) and emissions in conventional engines with 𝑅$^2$>0.91 and mean absolute errors (MAE) of less than five percent. Using the same approach, the hybrid vehicle state of charge (SOC), and the fuel scale state, are predicted, showing good agreement 𝑅$^2$>0.96 and confirming the versatility of the proposed algorithm. Finally, an initial approach for dealing with missing data in the databases is introduced. Using various simple and iterative imputation methods, it was possible to obtain 𝑅$^2$>0.80 for predicting the BTE and BSFC with five percent of the data missing from the input values.

The speech auditory brainstem response (sABR) is a promising technique for assessing the function of the auditory system. This non-invasive technique has shown utility as a marker of central processing disorders, some types of learning difficulties in children, and potentially for fitting hearing aids. However, the sABR needs a long recording time to obtain a reliable signal due to the high background noise, which limits its clinical applicability. The objective of this work is to develop methods to detect the sABR in high background noise and enhance it based on a modeling approach and through experimental testing. First, sABR noise estimation based on LQ/QR decomposition is derived, and its mathematical proof is shown. Second, an autoregression model is used to estimate the single-trial sABR which is then used to test several sABR detection and enhancement methods. Third, a novel Artificial Neural Network (ANN) based detection approach is proposed and compared using modeled and recorded data to other detection methods in the literature: Optimal Linear Filter (LF), Online Estimator (OE), Mutual Information (MI) and Artificial Neural Network based on the Discrete Wavelet Transform and Approximate Entropy (ANN DA). Finally, comprehensive evaluation of several sABR enhancement methods is performed, based on the Wiener Filter (WF), Maximum-SNR Filter (Max-SNR), Adaptive Noise Cancellation (ANC) with Least-Mean-Square (LMS), Affine Projection (AP) and Recursive-Least-Square (RLS) adaptation algorithm. The results show that the developed LQ/QR decomposition estimated noise is similar to the actual noise, and the modeled data are statistically similar to the recorded data. Moreover, the proposed ANN-based detection method is more accurate and requires less processing time than other methods, and the comprehensive evaluation of enhancement methods shows that RLS has best overall performance in enhancing the sABR. Therefore, the methods developed and evaluated in this work have the potential to reduce the required recording time for the sABR, and thus make it more practical as a clinical tool.

The first objective of this research project is to carry out multi-objective optimization (MOO) for four simple chemical engineering processes to clearly demonstrate the wealth of information on a given process that can be obtained from the MOO instead of a single aggregate objective function. The four optimization case studies are the design of a PI controller, an SO2 to SO3 reactor, a distillation column and an acrolein reactor. Results that were obtained from these optimization case studies show the benefit of generating and using the Pareto domain to gain a deeper understanding of the underlying relationships between the various process variables and the different performance objectives. In addition, an acrylic acid production plant model is developed in order to propose a methodology to solve multi-objective optimization for the two-reactor system model using artificial neural networks (ANNs) as metamodels, in an effort to reduce the computational time requirement that is usually very high when first-principles models are employed to approximate the Pareto domain. Once the metamodel was trained, the Pareto domain was circumscribed using a genetic algorithm and ranked with the Net Flow method (NFM). After the MOO was carry out with the ANN surrogate model, the optimization time was reduced by a factor of 15.5.