查看更多>>摘要：One of the most critical data analysis tasks is the streaming data classification, where we may also observe the concept drift phenomenon, i.e., changing the decision model's probabilistic characteristics. From a practical point of view, we may face this type of banking, medicine, or cybersecurity task to enumerate only a few. A vital characteristic of these problems is that the classes we are interested in (e.g., fraudulent transactions, treats, or serious diseases) are usually infrequent, which hinders the classification system design. The paper presents a novel algorithm DSCB (Deterministic Sampling Classifier with weighted Bagging) employs data preprocessing methods and weighted bagging technique to classify non-stationary imbalanced data stream. It builds models based on an incoming data chunk, but it also takes previously arrived instances into account. The proposed approach has been evaluated based on a wide range of computer experiments carried out on real and artificially generated data streams with various imbalance ratios, label noise levels, and concept drift types. The results confirmed that the weighted bagging ensemble coupled with data preprocessing could outperform state-of-the-art methods. (C)& nbsp;2022 Elsevier B.V. All rights reserved.
查看更多>>摘要：The present paper introduces Gesy, a genetic programming approach to script synthesis for zero-sum games. We will explore the sum-zero game context in Real-Time Strategy (RTS) games, where players must look for strategies (planning of actions) to maximize their gains or minimize their losses. The goal is to solve the script synthesis problem, which demands the synthesis of a computer program from a space of programs defined by a Domain-Specific Language (DSL). The synthesized program must encode a practical strategy for zero-sum games. Empirical results validate Gesy using the mu RTS platform, an academic test bed game that presents the main features found in RTS commercial games. The results show that our method provides interpretable strategies that are competitive with state-of-the-art search-based approaches in terms of play strength. Moreover, once synthesized, scripts require only a tiny fraction of the time needed by search-based methods to decide on the agent's next action. (c) 2022 Elsevier B.V. All rights reserved.
查看更多>>摘要：In this paper, two novel Multilayer Extreme Learning Machine (ML-ELM) networks are presented. We call them Improved Multilayer Extreme Learning Machines (IML-ELM). The proposed network architectures use neuron activations both during and after the training. In the first IML-ELM (IML-ELM1) network, each layer has connection weights assigned randomly as orthonormal. On the other hand, the second IML-ELM (IML-ELM2) has connection weights assigned randomly as orthonormal only in the first layer. Its following layers' connection weights are taken from the previous layer's output weight matrix. This assignment strategy made in the IML-ELM2 decreases the computation time even more. The networks' modeling performances on seven benchmark dynamic systems are investigated and it is shown that the proposed IML-ELM1 and IML-ELM2 perform better modeling than the ML-ELM. They have better modeling performance of more than 70% for both training and test data sets compared to ML-ELM for some systems studied. For instance, using 100 nodes, ML-ELM, IML-ELM1 and IML-ELM2 gave average testing root mean square error results of 0.627977, 0.104272 (83%) and 0.092683 (85%) respectively for BDS 7. In addition, it has been experimentally determined that the developed networks provide improvements in terms of average training time, and this improvement exceeds 60% in some cases. These achievements clearly prove that the proposed improved multilayer extreme learning machines are efficient tools for system modeling applications. (C) 2022 Elsevier B.V. All rights reserved.
查看更多>>摘要：Discernible patterns of a person's daily activities can be utilized to detect behavioral symptomatology of mental illness at early stages. Wearable Internet of Medical Things (IoMT) devices with sensors that collect motion data and provide objective measures of physical activity can help to better monitor and detect potential episodes related to the mental health conditions at earlier, more treatable stages. This research puts forward a neuro-symbolic model which uses learnable parameters with integrated knowledge for detection of depression episodes using IoMT based actigraphic input. A novel deep fuzzy model, Depress-DCNF is a hybrid of convolutional neural network (CNN) and an adaptive neuro fuzzy inference system (ANFIS) where CNN is used to extract high-level features from the motor activity recordings which are eventually combined with the discriminative statistical features to produce an optimized feature map. This optimized feature map is finally used to train the ANFIS model which accurately performs the depression classification task. The model is validated on the Depresjon benchmark dataset and compares favorably to state-of-the-art approach giving a superior performance accuracy of 85.10%.(C)& nbsp;2022 Elsevier B.V. All rights reserved.
查看更多>>摘要：The COrona VIrus Disease 2019 (COVID-19) pandemic is an ongoing global pandemic that has claimed millions of lives till date. Detecting COVID-19 and isolating affected patients at an early stage is crucial to contain its rapid spread. Although accurate, the primary viral test 'Reverse Transcription Polymerase Chain Reaction' (RT-PCR) for COVID-19 diagnosis has an elaborate test kit, and the turnaround time is high. This has motivated the research community to develop CXR based automated COVID-19 diagnostic methodologies. However, COVID-19 being a novel disease, there is no annotated large-scale CXR dataset for this particular disease. To address the issue of limited data, we propose to exploit a large-scale CXR dataset collected in the pre-COVID era and train a deep neural network in a self supervised fashion to extract CXR specific features. Further, we compute attention maps between the global and the local features of the backbone convolutional network while finetuning using a limited COVID-19 CXR dataset. We empirically demonstrate the effectiveness of the proposed method. We provide a thorough ablation study to understand the effect of each proposed component. Finally, we provide visualizations highlighting the critical patches instrumental to the predictive decision made by our model. These saliency maps are not only a stepping stone towards explainable AI but also aids radiologists in localizing the infected area. (C)& nbsp;2022 Elsevier B.V. All rights reserved.
查看更多>>摘要：In this paper we have addressed the extraction of hidden knowledge from medical records using data mining techniques such as association rules in conjunction with fuzzy logic in a distributed environment. A significant challenge in this domain is that although there are a lot of studies devoted to analysing health data, very few focus on the understanding and interpretability of the data and the hidden patterns present within the data. A major challenge in this area is that many health data analysis studies have focussed on classification, prediction or knowledge extraction and end users find little interpretability or understanding of the results. This is due to the use of black-box algorithms or because the nature of the data is not represented correctly. This is why it is necessary to focus the analysis not only on knowledge extraction but also on the transformation and processing of the data to improve the modelling of the nature of the data. Techniques such as association rule mining and fuzzy logic help to improve the interpretability of the data and treat it with the inherent uncertainty of real-world data. To this end, we propose a system that automatically: a) pre-processes the database by transforming and adapting the data for the data mining process and enriching the data to generate more interesting patterns, b) performs the fuzzification of the medical database to represent and analyse real-world medical data with its inherent uncertainty, c) discovers interrelations and patterns amongst different features (diagnostic, hospital discharge, etc.), and d) visualizes the obtained results efficiently to facilitate the analysis and improve the interpretability of the information extracted. Our proposed system yields a significant increase in the compression and interpretability of medical data for end-users, allowing them to analyse the data correctly and make the right decisions. We present one practical case using two health-related datasets to demonstrate the feasibility of our proposal for real data. (c) 2022 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
查看更多>>摘要：This paper addresses the electric demand prediction problem using neural networks and symbolization techniques. Symbolization techniques provide a time series symbolic representation of a lower length than the original time series. In our methodology, we incorporate the use of encoding from ordinal regression, preserving the notation of order between the symbols and make extensive experimentation with different neural network architectures and symbolization techniques. In our experimentation, we used the total electric demand data in the Spanish peninsula electric network, taken from 2009 to 2019 with a granularity of 10 min. The best model found making use of the symbolization methodology offered us slightly worse quality metrics (1.3655 RMSE and 0.0390 MAPE instead of the 1.2889 RMSE and 0.0363 MAPE from the best numerical model) but it was trained 6826 times faster. (C) 2022 Elsevier B.V. All reserved.
查看更多>>摘要：This paper proposes a novel decentralize and asynchronous swarm robotic search algorithm integrated with game theory to better disperse robots in the environment while crossing obstacles and solving mazes. This prevents early convergence and improves the efficiency of the searches. In the proposed algorithm, individual robots, while searching, play a sequential game at each iteration, and based on that, choose their velocity update rule. The effectiveness of the proposed strategic game is tested in a specially designed framework. As a validation, the introduced algorithm is compared with the state-of-the-art in simple and complex search environments. The results showed that the suggested algorithm outperforms other methods both in search duration and attained path length to the target, and its success rate is equal to the one of state-of-the-art (i.e., 100% in the conducted experiments). Also, it is shown that the proposed strategic game works well in search environments with different levels of complexity and especially improves search efficiency further in complex environments. (C) 2022 Elsevier B.V. All rights reserved.
查看更多>>摘要：The current literature on water demand forecasting mostly focuses on giving accurate point predictions of water demand. However, the water demand point forecasting will encounter uninformative and unreliable problems when the uncertainty level of data increases. To solve the above problem, a hybrid model (KDE-PSO-LSTM), which combines long short-term memory networks (LSTM) to kernel density estimation (KDE) optimized by using the particle swarm optimization (PSO) algorithm, is proposed to acquire the water demand prediction interval (PI) to quantify the likely uncertainties in the predictions. At first, the prediction errors are obtained by the difference between the real values of water demand and the predictive values based on the LSTM model. Then, a novel splitting strategy is proposed to divided point predictions into different levels to deal with the problem that it is difficult to fit the prediction errors of the whole water demand using a single probability density function (PDF). Next, the PSO is used to optimize the hyper-parameter of the KDE method for fitting the PDF curves of different levels prediction errors. Moreover, due to the irregular distribution of prediction errors, a search method called confidence-window shifting is presented to determine the optimal prediction error interval from the fitted PDF curves. After that, the upper bounds and the lower bounds of the best intervals of prediction errors are added to the point predictions to attain the final PI of urban water demand. Finally, to demonstrate the superiorities of the proposed model, the proposed KDE-PSO distribution is compared to other well-known distributions, i.e, the KDE distribution, the Beta-PSO distribution and the normal distribution. The experimental results show that the comprehensive performances of the PIs generated from the proposed KDE-PSO-LSTM model are better than that of KDE-PSO-BP, KDE-PSO-RNN, ND-LSTM, KDE-LSTM, Beta-PSO-LSTM and KDE-GA-LSTM. Therefore, it can be demonstrated that the KDE-PSO-LSTM model can provide reliable decision support to policy-makers for making the optimal water supplying management. (C) 2022 Elsevier B.V. All rights reserved.
查看更多>>摘要：An accurate electrical load forecasting is essential for optimal grid operation. The paper presents a methodology for the short-term commercial building electrical load forecasting through a regularized deep neural network: Long Short-Term Memory Recurrent Neural Network (LSTM-RNN). Detailed heuristic analysis regarding relevant input feature selection, the volume of training data, hyperparameter tuning and regularizer selection of an optimal LSTM-RNN network configuration is presented. The regularized LSTM-RNN is used to forecast 30-min and 24-h ahead electrical loads of two commercial buildings in Virginia, USA. The forecast is performed for one week each over four different months in 2019: January, April, July and October to represent four different seasons in North America. The performance of electrical load forecasts has been compared against actual smart meter data from the electric utility of these buildings. For the case study presented, Mean Absolute Percentage Error (MAPE%) with the regularized LSTM-RNN is 4.9%, compared to 6.4%, 9.2% and 13.3% with Shallow-ANN (Artificial Neural Network), Support Vector Regression (SVR) and Linear Regression (LR) respectively for 30-min ahead electrical load forecast. For 24-h ahead electrical load forecast, MAPE (%) is 11.6%, compared to 12.7%, 13.4% and 14.3% with shallow-ANN, SVR and LR respectively. The methodology to configure a deep neural network (LSTM-RNN) for electrical load forecasting presented in this paper can be utilized for optimal forecasting performance. (C) 2022 Elsevier B.V. All rights reserved.
NSTL
Elsevier