Search

It is widely recognised that learning systems have to go deeper to exchange for more powerful representation learning capabilities in order to precisely approximate nonlinear complex problems. However, the best known computational intelligence approaches with such characteristics, namely, deep neural networks, are often criticised for lacking transparency. In this paper, a novel multilayer evolving fuzzy neural network (MEFNN) with a transparent system structure is proposed. The proposed MEFNN is a meta-level stacking ensemble learning system composed of multiple cascading evolving neuro-fuzzy inference systems (ENFISs), processing input data layer-by-layer to automatically learn multi-level nonlinear distributed repre- sentations from data. Each ENFIS is an evolving fuzzy system capable of learning from new data sample by sample to self- organise a set of human-interpretable IF-THEN fuzzy rules that facilitate approximate reasoning. Adopting ENFIS as its ensemble component, the multilayer system structure of MEFNN is flexible and transparent, and its internal reasoning and decision-making mechanism can be explained and interpreted to/by humans. To facilitate information exchange between different layers and at- tain stronger representation learning capability, MEFNN utilises error backpropagation to self-update the consequent parameters of the IF-THEN rules of each ensemble component based on the approximation error propagated backward. To enhance the capability of MEFNN to handle complex problems, a nonlinear activation function is introduced to modelling the consequent parts of the IF-THEN rules of ENFISs, thereby empowering both the representation and the reflection of nonlinearity in the resulting fuzzy outputs. Numerical examples on a wide variety of challenging (benchmark and real-world) classification and regression problems demonstrate the superior practical performance of MEFNN, revealing the effectiveness and validity of the proposed approach.

This study proposes to construct a model using random forest method, an efficient machine learning-based method, to predict the spatial structure and temporal evolution of the sea surface temperature (SST) cooling induced by northwest Pacific tropical cyclones (TCs), a process of the so-called wind pump. The predictors in use include 12 predictors related to TC characteristics and pre-storm ocean conditions. The model is shown to skillfully predict the spatiotemporal evolutions of the cold wake generated by TCs of different intensity groups, and capture the cross-case variance in the observed SST response. Another model is further built based on the same method to assess the relative importance of the 12 predictors in determining the magnitude of the maximum cooling. Computations of feature scores of those predictors show that TC intensity, translation speed and size, and pre-storm mixed layer depth and SST dominate, depending on the area where the cooling is considered.

Sea surface height anomaly (SSHA) induced by tropical cyclones (TCs) is closely associated with oscillations and is a crucial proxy for thermocline structure and ocean heat content in the upper ocean. The prediction of TC-induced SSHA, however, has been rarely investigated. This study presents a new composite analysis-based random forest (RF) approach to predict daily TC wind pump induced SSHA. The proposed method utilizes TC's characteristics and prestorm upper oceanic parameters as input features to predict TC-induced SSHA up to 30 days after TC passage. Simulation results suggest that the proposed method is skillful at inferring both the amplitude and temporal evolution of SSHA induced by TCs of different intensity groups. Using a TC-centered 5° × 5° box, the proposed method achieves highly accurate prediction of TC-induced SSHA over the Western North Pacific with root mean square error of 0.024 m, outperforming alternative machine learning methods and the numerical model. Moreover, the proposed method also demonstrated good prediction performance in different geographical regions, i.e., the South China Sea and the Western North Pacific subtropical ocean. The study provides insight into the application of machine learning in improving the prediction of SSHA influenced by extreme weather conditions. Accurate prediction of TC-induced SSHA allows for better preparedness and response, reducing the impact of extreme events (e.g., storm surge) on people and property.

Evolving fuzzy systems (EFSs) are widely known as a powerful tool for streaming data prediction. In this paper, a novel zero-order EFS with a unique belief structure is proposed for data stream classification. Thanks to this new belief structure, the proposed model can handle the inter-class overlaps in a natural way and better capture the underlying multi-model structure of data streams in the form of prototypes. Utilizing data-driven soft thresholds, the proposed model self-organizes a set of prototype-based IF-THEN fuzzy belief rules from data streams for classification, and its learning outcomes are practically meaningful. With no requirement of prior knowledge in the problem domain, the proposed model is capable of self-determining the appropriate level of granularity for rule base construction, while enabling users to specify their preferences on the degree of fineness of its knowledge base. Numerical examples demonstrate the superior performance of the proposed model on a wide range of stationary and nonstationary classification benchmark problems.

Adaptive boosting (AdaBoost) is a widely used technique to construct a stronger ensemble classifier by combining a set of weaker ones. Zero-order fuzzy inference systems (FISs) are very powerful prototype-based predictive models for classification, offering both great prediction precision and high user-interpretability. However, the use of zero-order FISs as base classifiers in AdaBoost has not been explored yet. To bridge the gap, in this paper, a novel multi-class fuzzily weighted AdaBoost (FWAdaBoost)-based ensemble system with self-organising fuzzy inference system (SOFIS) as the ensemble component is proposed. To better incorporate SOFIS, FWAdaBoost utilises the confidence scores produced by SOFIS in both sample weight updating and ensemble output generation, resulting in more accurate classification boundaries and greater prediction precision. Numerical examples on a wide range of benchmark classification problems demonstrate the efficacy of the proposed approach.

Large-scale (large-area), fine spatial resolution satellite sensor images are valuable data sources for Earth observation while not yet fully exploited by research communities for practical applications. Often, such images exhibit highly complex geometrical structures and spatial patterns, and distinctive characteristics of multiple land-use categories may appear at the same region. Autonomous information extraction from these images is essential in the field of pattern recognition within remote sensing, but this task is extremely challenging due to the spectral and spatial complexity captured in satellite sensor imagery. In this research, a semi-supervised deep rule-based approach for satellite sensor image analysis (SeRBIA) is proposed, where large-scale satellite sensor images are analysed autonomously and classified into detailed land-use categories. Using an ensemble feature descriptor derived from pre-trained AlexNet and VGG-VD-16 models, SeRBIA is capable of learning continuously from both labelled and unlabelled images through self-adaptation without human involvement or intervention. Extensive numerical experiments were conducted on both benchmark datasets and real-world satellite sensor images to comprehensively test the validity and effectiveness of the proposed method. The novel information mining technique developed here can be applied to analyse large-scale satellite sensor images with high accuracy and interpretability, across a wide range of real-world applications.

Aerial scene classification is the key task for automated aerial image understanding and information extraction, but is highly challenging due to the great complexity and real-world uncertainties exhibited by such images. To perform precise aerial scene classification, in this research, a multistream deep rule-based ensemble system is proposed. The proposed ensemble system consists of three deep rule-based systems that are trained simultaneously on the same data. The three ensemble components employ ResNet50, DenseNet121, and InceptionV3 as their respective feature descriptors because of the state-of-the-art performances the three networks have demonstrated on aerial scene classification. The three networks are fine-tuned on aerial images to further enhance their discriminative and descriptive abilities. Thanks to its prototype-based nature, the proposed approach is able to self-organize a transparent ensemble predictive model with prototypes learned from training images and perform highly explainable joint decision-making on testing images with greater precision. Numerical examples based on both benchmark aerial image sets and satellite sensor images demonstrated the efficacy of the proposed approach, showing its great potential in solving real-world problems.

Adaptive boosting (AdaBoost) is a widely used technique to construct a stronger ensemble classifier by combining a set of weaker ones. Zero-order fuzzy inference systems (FISs) are very powerful prototype-based predictive models for classification, offering both great prediction precision and high user interpretability. However, the use of zero-order FISs as base classifiers in AdaBoost has not been explored yet. To bridge the gap, in this article, a novel multiclass fuzzily weighted AdaBoost (FWAdaBoost)-based ensemble system with a self-organizing fuzzy inference system (SOFIS) as the ensemble component is proposed. To better incorporate the SOFIS, FWAdaBoost utilizes the confidence scores produced by the SOFIS in both sample weight updating and ensemble output generation, resulting in more accurate classification boundaries and greater prediction precision. Numerical examples on a wide range of benchmark classification problems demonstrate the efficacy of the proposed approach.