Expert Systems with Applications: X最新文献

This paper proposes a filter-based feature selection method by combining the measurement of kernel canonical correlation analysis (KCCA) with the mutual information (MI)-based feature selection method, named mRMJR-KCCA. The mRMJR-KCCA maximizes the relevance between the feature candidate and the target class labels and simultaneously minimizes the joint redundancy between the feature candidate and the already selected features in the view of KCCA. To improve the computation efficiency, we adopt the Incomplete Cholesky Decomposition to approximate the kernel matrix in implementing the KCCA in mRMJR-KCCA for larger-size datasets. The proposed method is experimentally evaluated on 13 classification-associated datasets. Compared with certain popular feature selection methods, the experimental results demonstrate the better performance of the proposed mRMJR-KCCA.

Forecasting future trends of the stock market using the historical data is the exigent demand in the field of academia as well as business. This work has explored the feature optimization capacity of firefly with an evolutionary framework considering the biochemical and social aspects of Firefly algorithm, along with the selection procedure of objective value in evolutionary notion. The performance of the proposed model is evaluated using four different stock market datasets, such as BSE Sensex, NSE Sensex, S&P 500 index and FTSE index. The datasets are regenerated using the proper mathematical formulation of the fundamental part belonging to technical analysis, such as technical indicators and statistical measures. The feature reduction through transformation is carried out on the enhanced dataset before employing the experimented dataset to the prediction models such as Extreme Learning Machine (ELM), Online Sequential Extreme Learning Machine (OSELM) and Recurrent Back Propagation Neural Network (RBPNN). For feature reduction, both statistical and optimized based feature reduction strategies are considered, where Principal Component Analysis (PCA) and Factor Analysis (FA) are examined for statistical based feature reduction and Firefly Optimization (FO), Genetic Algorithm (GA) and Firefly algorithm with evolutionary framework are well thought out for optimized feature reduction techniques. An empirical comparison is established among the experimented prediction models considering all the feature reduction techniques for the time horizon of 1 day, 3 days, 5 days, 7 days, 5 days and 30 days in advance, applying on all the datasets used in this study. From the simulation result, it can be clearly figured out that firefly with evolutionary framework optimized feature reduction applying to OSELM prediction model outperformed over the rest experimented models.

Due to the power and impact of social media, unsolved practical issues such as human trafficking, kinship recognition, and clustering family photos from large collections have recently received special attention from researchers. In this paper, we present a new idea for family and non-family photo classification. Unlike existing methods that explore face recognition and biometric features, the proposed method explores the strengths of facial geometric features and texture given by a new fractional-entropy approach for classification. The geometric features include spatial and angle information of facial key points, which give spatial and directional coherence. The texture features extract regular patterns in images. The proposed method then combines the above properties in a new way for classifying family and non-family photos with the help of Convolutional Neural Networks (CNNs). Experimental results on our own as well as benchmark datasets show that the proposed approach outperforms the state-of-the-art methods in terms of classification rate.

The authors present an expert and intelligent system that (1) identifies influential term groups having causal relationships with real-world enterprise outcomes from Twitter data and (2) quantifies the appropriate time lags between identified influential term groups and enterprise outcomes. Existing expert and intelligent systems, which are defined as computer systems that imitate the ability of human decision making, could enable computers to identify the spread of Twitter users’ enterprise-related feedback automatically. However, existing expert and intelligent systems have limitations on automatically identifying the causal effects on enterprise outcomes. Identifying the causal effects on enterprise outcomes is important, because Twitter users’ feedback toward enterprise decisions may have real-world implications. The proposed expert and intelligent system can support decision makers’ decisions considering the real-world effects of identified Twitter users’ feedback on enterprise outcomes. In particular, (1) a co-occurrence network analysis model is exploited to discover term candidates for generating influential term groups that are combinations of enterprise-related terms, which potentially influence enterprise outcomes. (2) Time series models and (3) a Granger causality analysis model are then employed to identify influential term groups having causal relationships with enterprise outcomes with the appropriate time lags. Case studies involving a real-world internet video streaming and disc rental provider as well as an airline company are used to test the validity of the proposed expert and intelligent system for both predicting enterprise outcomes in a long period and predicting the effects of specific events on enterprise outcomes in a short period.

The Publisher regrets that this article is an accidental duplication of an article that has already been published in Expert Systems with Applications, volume 152, 15 August 2020, 113382 http://dx.doi.org/10.1016/j.eswa.2020.113382. The duplicate article has therefore been withdrawn.

The full Elsevier Policy on Article Withdrawal can be found at http://www.elsevier.com/locate/withdrawalpolicy.

The layout of manufacturing facilities has a large impact on manufacturing performance. The layout design process produces a block plan that shows the relative positioning of resources that can be developed into a detailed layout drawing. The total material handling distance is commonly used for measuring material flow. Manufacturing systems are subject to external and internal uncertainties including demand and machine breakdowns. Uncertainty and the rerouting of material flows have an impact on the material handling distance. No previous research has integrated robust machine layout design through multiple periods of dynamic demand with machine maintenance planning. This paper presents a robust machine layout design tool that minimises the material flow distance using a Genetic Algorithm (GA), taking into account demand uncertainty and machine maintenance. Experiments were conducted using eleven benchmark datasets that considered three scenarios: preventive maintenance (PM), corrective maintenance (CM) and both PM and CM. The results were analysed statistically. The effect of several maintenance scenarios including the ratio of the number of machines with period-based PM (PPM) to the number with production quantity-based PM (QPM), the percentage of machines with CM (%CM), and a combination of PMM/QPM ratios and %CM on material flow distance were examined. The results show that designing robust layouts considering maintenance resulted in shorter material flow distances. The distance was decreased by 30.91%, 9.8%, and 20.7% for the PM, CM, and both PM/CM scenarios, respectively. The PPM/QPM ratios, %CM, and a combination of PPM/QPM and %CM had significantly resulted in the material flow distance on almost all datasets.

The effective prediction of pavement performance trends can help in achieving the cost-effective management of pavements over their service life. The international roughness index (IRI) is a widely used pavement performance index, which can be considered as a time-dependent variable in terms of scientific modeling. This research aims to develop an innovative IRI prediction model based on fuzzy-trend time-series forecasting and particle swarm optimization (PSO) techniques. Raw datasets extracted from the Long-Term Pavement Performance database are used for model training, testing, and performance assessment. First, IRI values are divided into different granular spaces, which are considered as the principal factor and subfactors. In addition, the multifactor interval division method is proposed according to the principle of the automatic clustering technique. Next, a second-order fuzzy-trend model and fuzzy-trend relationship classification method are proposed to predict the fuzzy-trend of each factor. Then, the fuzzy-trend states for multiple granular spaces are generated while giving full consideration to various uncertainties. Finally, the PSO technique is used to optimize the performance model while carrying out future IRI forecasting. Comparative experiments are performed using more than 20,000 data items from different regions to verify the effectiveness of the proposed method. The experimental results indicate that the proposed method outperforms other approaches including the polynomial fitting, autoregressive integrated moving average, and backpropagation neural network methods in terms of the root mean squared error (0.191) and relative error (6.37%).

Context

Recent developments in requirements engineering (RE) methods have seen a surge in using machine-learning (ML) algorithms to solve some difficult RE problems. One such problem is identification and classification of non-functional requirements (NFRs) in requirements documents. ML-based approaches to this problem have shown to produce promising results, better than those produced by traditional natural language processing (NLP) approaches. Yet, a systematic understanding of these ML approaches is still lacking.

Method

This article reports on a systematic review of 24 ML-based approaches for identifying and classifying NFRs. Directed by three research questions, this article aims to understand what ML algorithms are used in these approaches, how these algorithms work and how they are evaluated.

Results

(1) 16 different ML algorithms are found in these approaches; of which supervised learning algorithms are most popular. (2) All 24 approaches have followed a standard process in identifying and classifying NFRs. (3) Precision and recall are the most used matrices to measure the performance of these approaches.

Finding

The review finds that while ML-based approaches have the potential in the classification and identification of NFRs, they face some open challenges that will affect their performance and practical application.

Impact

The review calls for the close collaboration between RE and ML researchers, to address open challenges facing the development of real-world ML systems.

Significance

The use of ML in RE opens up exciting opportunities to develop novel expert and intelligent systems to support RE tasks and processes. This implies that RE is being transformed into an application of modern expert systems.

In this paper, a novel multi-module neural network system named MMNNS is proposed to solve the imbalance problem in electrocardiogram (ECG) heartbeats classification. Four submodules are designed to construct the system: preprocessing, imbalance problem processing, feature extraction and classification. Imbalance problem processing module mainly introduces three methods: BLSM, CTFM and 2PT, which are proposed from three aspects of resampling, data feature and algorithm respectively. BLSM is used to synthesize virtual samples linearly around the minority samples. CTFM consists of DAE-based feature extraction part and QRS-based feature selection part, in which selected features and complete features are applied to determine the heartbeat class simultaneously. The processed data are fed into a convolutional neural network (CNN) by applying 2PT to train and fine-tune. MMNNS is trained on MIT-BIH Arrhythmia Database following AAMI standard, using intra-patient and inter-patient scheme, especially the latter which is strongly recommended. The comparisons with several state-of-the-art methods using standard criteria on three datasets demonstrate the superiority of MMNNS for improving detection of heartbeats and addressing imbalance in ECG heartbeats classification.