Selecting risk response strategies to minimize human errors in a design project for factories of the future (2023)

Expert Systems with Applications, Volume 225, 2023, Article 120138

Random Forest is an ensemble classification approach. It aims to design a discrete finite group of decision trees constructed based on bootstrap samples and random attribute selection. Random Forests have strong generalization capacities due to the variance in the training and attribute couple subsets used for constructing different decision trees in the forest. However, to construct a robust and effective random forest, two main issues need to be taken into account namely: (1) increasing the accuracy and diversity of decision trees; (2) decreasing the number of decision trees. In this paper, a genetic algorithm-based approach to tackle the aforementioned challenges related to random forest construction is proposed. Three objectives are taken into consideration. First, strengthening the classification accuracy of individual decision trees as well as that of the forest. Second, making use of diversity measures among the decision trees to improve the generalization of the constructed model. Third, minimizing the number of trees in the forest and finding an optimal subset of the random forest. An experimental evaluation on several datasets from the UCI Machine Learning Repository is conducted. The obtained results show that the proposed approach outperforms state-of-the-art classical as well as evolutionary random forest construction methods. Finally, the proposed approach is used to build a reliable random forest model for detecting Botnet traffic in Internet of Things environment.

Expert Systems with Applications, Volume 225, 2023, Article 120090

Many urban road networks today are experiencing increasing congestion that threatens not only transport efficiency but also living environment. To solve these problems, providing proactive knowledge of traffic performance is of significant importance. However, due to the inherent uncertainties of the signalized urban road network, it is a challenging work and some gaps still exist. First, the existing approaches are usually limited to a single location or region. Second, traffic flow parameters such as volume are used as a proxy of traffic state. Aiming to fill these gaps, this study developed a dynamic factor model-based approach to forecast the multi-step network traffic states of a group of regions in urban road network collaboratively. The novel model decomposes a set of traffic state time series into two orthogonal components: the common latent factor and the idiosyncratic disturbance. The common latent factor drives the co-movement dynamics of network traffic states, while the idiosyncratic disturbance captures the region-specific distinctions of traffic states in different regions. By extracting the principal variations of traffic states in a group of regions, the proposed model reduces the dimensionality of the traffic state variable from high-dimensional original space to low-dimensional latent factor space. By means of the common latent factor and its evolution over time, spatial–temporal correlations of traffic states in different regions and different time slots are seamlessly incorporated. The proposed model exhibits four distinct advantages, (1) it collaboratively produces forecasts for a group of regions; (2) it considers both co-movement dynamics and region-specific distinctions of traffic state; (3) the model incorporates both spatial and temporal correlations seamlessly, and (4) it reduces dimensionality from a network-wide high-dimensional space to a low-dimensional latent factors space. The proposed model is applied to the real urban road network in Shanghai, China, based on the large-scale traffic performance index data released by the Shanghai Transportation Big Data Joint Innovation Laboratory. Empirical results from extensive experiments demonstrate the proposed dynamic factor model provides a promising approach for multi-step network traffic state forecast in urban road network, and outperforms the competing models considered in this study.

Expert Systems with Applications, Volume 225, 2023, Article 120210

The task of biomedical question answering is a subtask of the more general question answering task, that is concerned only with biomedical questions. The current state-of-the-art models in this task like BioBERT, and BioM-ELECTRA are all based on the transformer architecture. The self-attention layer in the transformer plays a central role in the model predictions. Recent studies on the inner-workings of the transformer self-attention layer in the case of question answering hypothesize that context passage tokens with bigger attention scores have a bigger probability of being part of the predicted answer. Starting from this hypothesis, we experimented with a novel self-attention enriching mechanism for biomedical question answering targeting factoid and list type questions. In our approach, we enrich BioBERT’s self-attention layer with biomedical and named entity information previously extracted from the question and the context passage. The proposed enriching mechanism increases the attention scores for the biomedical and named entities. Which are in most cases the answer to the question. This increase in attention scores influences the model final prediction as hypothesized. Our proposed method achieves state-of-the-art results on several batches of the BioASQ’s 10b, 9b, 8b, and 7b datasets.

Expert Systems with Applications, Volume 225, 2023, Article 120109

The advancement of smart meters using evolving technologies such as the Internet of Things (IoT) is producing more data for the training of energy prediction models. Since many machine learning techniques were not premeditated to handle a large number of irrelevant features, it has engendered the search for optimal techniques to decrease the generated features and potentially identify the most relevant features that have an impact on building energy efficiency. Feature selection is considered one of the most suitable methods of pinpointing the best features combination. However, the fraction of studies that deliver comprehensive insights on the incorporation of feature selection with machine learning is still limited, notwithstanding the capabilities of feature selection to produce a good result in terms of accuracy and speed. To address this gap, this study investigates feature selection methods centred on building energy consumption prediction using machine learning. This study conducted a comparative analysis of 14 machine learning algorithms on 5 different data sizes and explored the effect of 7 feature selection methods on model performance for predicting energy consumption in buildings. Furthermore, this study identifies the most effective feature selection methods and machine learning models for energy use prediction. The experimental results demonstrate thatfeature selection can affect model’s performance positively or negatively, depending on the algorithm employed. Nevertheless, the filter methodwas noted as the most appropriate methodfor most Machine Learning (ML) classification algorithms. Moreover, Gradient Boosting (GB) was identified as the most effective modelfor predicting energy performance in buildings.Additionally, the reliability analysis confirms the hypothesis that “thelarger the data, the more accurate the result” but only for specific algorithms such as Deep Neural Networks (DNN).This study also presents the theoretical and practical implications of this research.

Expert Systems with Applications, Volume 225, 2023, Article 120088

Crowd counting in adverse hazy weather is inevitable and significant to the scene understanding in real-world application. For the crucial and challenging hazy-weather crowd counting issue, this paper proposes a single-stage hazy-weather crowd counting method based on direction-aware attention aggregation. The method develops a Directional Context Attention (DCA) block to realize embedding the positional information to the channel attention aggregation and to capture the long-range dependencies about the crowd structure, both of which guide the counting method readily locate the crowd of interest. Due to the lack of the hazy-weather crowd counting benchmark datasets, we also generate the hazy-weather crowd counting datasets to evaluate the proposed method. Experimental results and the discussions demonstrate our method is effective greatly on the hazy-weather crowd counting task, especially the design of the core DCA block. In this research work, both the proposed method and the generated benchmark datasets in hazy-weather scene facilitate the development of the crowd counting in the adverse weather conditions and the universal intelligent surveillance technologies. The generated hazy-weather crowd counting benchmarks datasets (Hazy-ShanghaiTechRGBD and Hazy-JHU) in this paper would be released in https://github.com/312524/Hazy-CC after peer-review process of this work to facilitate further research.

Expert Systems with Applications, Volume 225, 2023, Article 120064

The popularity of intelligent mobile devices has increased because they are not only convenient for people but also produce a large number of GPS trajectories. Semantic trajectories can be obtained by adding semantic information such as landmarks and activities to raw trajectories. Keyword queries in semantic trajectory databases that return the relevant places/routes have attracted increasing attention from researchers in recent years. However, existing works only consider the spatial and textual features of keywords, which cannot answer queries with temporal requirements. Simply modifying existing algorithms to support temporal requirements may lead to errors and low efficiency. Additionally, they match keywords only by string similarity without considering their semantic meanings. In this paper, we study the problem of efficient spatiotemporal keyword search in semantic trajectories (STKST). Given the position of a user and a set of keywords with temporal constraints, we aim to efficiently retrieve top-$k$ trajectories that contain the most semantically and temporally relevant keywords and are close to the position of the user. To measure the goodness of a trajectory regarding the query, we devise a new integrated similarity measure by considering information from three aspects (spatial, temporal, and semantic). Then we develop a novel hybrid spatial–temporal–semantic index (STS-I) to organize these three kinds of information in trajectories in the form of tree structure. Finally, we propose a new algorithm STKST-I to efficiently prune unqualified trajectories based on the lower and upper bounds derived from the STS-I index. Extensive experimental studies are conducted on real trajectory datasets to verify the performance of our methods.