Stationary Data Sets Research Articles

Fast online feature selection in streaming data

The challenge of getting big amounts of high-quality labeled data is compounded by the fact that data labeling is often subjective and requires significant human effort. In many cases, the quality of the labeled data depends entirely on the expertise and experience of human annotators, making it challenging to ensure labeling accuracy in large and dynamic datasets. Moreover, there may be a significant delay between the arrival of a new instance and its manual labeling. This paper explores the use of fully unsupervised feature selection algorithms in non-stationary data streams, where the importance of features may change over time. We introduce a novel feature selection algorithm called Online Fast FEa-ture SELection-OFFESEL, which calculates the feature importance scores in each incoming window based on their mean normalized values and without using any class labels. We evaluate OFFESEL on 17 benchmark data streams, both stationary and non-stationary, using popular online classifiers like PerceptronMask, VFDT, Online Boosting, and Linear SVM. We compare OFFESEL to several other feature selection algorithms, including state-of-the-art supervised ones like FIRES and ABFS, as well as popular unsupervised ones like MCFS, LS, and Max Variance, which we adapted to data streams. Our results indicate that OFFESEL outperforms all supervised and unsupervised feature selection algorithms in terms of classification accuracy. Specifically, OFFESEL preserves the accuracy level of the supervised FIRES algorithm, which proved more accurate than ABFS in our experiments, while maintaining the accuracy level achieved by the unsupervised Max Variance algorithm. Moreover, OFFESEL requires even less computation time than Max Variance and shows high stability on stationary datasets. Overall, our study demonstrates the potential benefits of using unlabeled data for feature ranking and selection in dynamic data streams.

Modeling Urban Temperature Using Measurements from Mobile and Stationary Monitoring Stations

Heat waves are occurring more frequently worldwide as global warming continues, and urban heat islands can threaten conventional life in cities. Measuring, analyzing, and simulating weather data at fine spatial and temporal scales are essential to prevent and reduce the damage caused by extreme heat waves. In urban environments, handling complex micrometeorological situations using current meteorological stations and global simulation models (e.g., weather research forecasting models) is challenging. In this study, the thermal environments of urban areas were measured using a mobile meteorological measurement platform. Both mobile and stationary datasets were incorporated into the meteorological modeling process to simulate the spatial and temporal distribution of temperature. Additionally, various mobile observation implementation scenarios for temperature modeling were examined. We compared simulation combinations with the temperature field generated from the total dataset to obtain a better sampling campaign and properly incorporate mobile data scenarios. When collecting mobile data, it is important to consider spatial features to improve the efficiency of sampling programs. This can substantially reduce the cost of mobile data collection, together with the sensor error bound.

Framework for metamodel-based design optimization considering product performance and assembly process complexity

This paper proposes a method for simultaneous evaluation of the assembly process complexity together with the performance of the future product. It allows for product design optimization, considering different aspects of the future design at the early stage of the development process. The proposed method, embodied in a fully automated framework, substitutes the traditional sequential development process with a more efficient and rapid combined procedure, which addresses multiple design aspects simultaneously. Design for assembly (DFA) rules, used as quantitative metrics of the ease-of-assembly of the whole product and individual assembly operations, are automatically evaluated together with performance metrics, estimated based on finite element (FE) simulations. The direct solution to this optimization problem might be inefficient or impossible since it requires the recurrent evaluation of computationally expensive discrete and continuous functions with unknown behavior that represent the optimization objectives and constraints. For that reason, the proposed framework employs regression models based on the Gaussian process and artificial neural networks, thus achieving the optimal design of a product as a result of metamodel-based design optimization (MBDO). The suggested approach is demonstrated in the optimization of a gearbox assembly, considering its mechanical performance and assembly process. Comparing the results of the metamodel-based and direct design optimization shows that MBDO allows finding a better solution using a three times smaller computational budget. In addition, analysis of the results obtained using stationary sampling data sets of different sizes highlighted the limitations of the employed sampling procedure.

Brain-Inspired Fast-and Slow-Update Prompt Tuning for Few-Shot Class-Incremental Learning.

Few-shot class-incremental learning (FSCIL) aims to learn new classes incrementally with a limited number of samples per class. Foundation models combined with prompt tuning showcase robust generalization and zero-shot learning (ZSL) capabilities, endowing them with potential advantages in transfer capabilities for FSCIL. However, existing prompt tuning methods excel in optimizing for stationary datasets, diverging from the inherent sequential nature in the FSCIL paradigm. To address this issue, taking inspiration from the "fast and slow mechanism" of the complementary learning systems (CLSs) in the brain, we present fast-and slow-update prompt tuning FSCIL (FSPT-FSCIL), a brain-inspired prompt tuning method for transferring foundation models to the FSCIL task. We categorize the prompts into two groups: fast-update prompts and slow-update prompts, which are interactively trained through meta-learning. Fast-update prompts aim to learn new knowledge within a limited number of iterations, while slow-update prompts serve as meta-knowledge and aim to strike a balance between rapid learning and avoiding catastrophic forgetting. Through experiments on multiple benchmark tests, we demonstrate the effectiveness and superiority of FSPT-FSCIL. The code is available at https://github.com/qihangran/FSPT-FSCIL.

Multi-Scale Batch-Learning Growing Neural Gas Efficiently for Dynamic Data Distributions

Growing neural gas (GNG) has many applications, including topology preservation, feature extraction, dynamic adaptation, clustering, and dimensionality reduction. These methods have broad applicability in extracting the topological structure of 3D point clouds, enabling unsupervised motion estimation, and depicting objects within a scene. Furthermore, multi-scale batch-learning GNG (MS-BL-GNG) has improved learning convergence. However, it is only implemented on static or stationary datasets, and adapting to dynamic data remains difficult. Similarly, the learning rate cannot be increased if new nodes are added to the existing network after accumulating errors in the sampling data. Next, we propose a new growth approach that, when applied to MS-BL-GNG, significantly increases the learning speed and adaptability of dynamic data distribution input patterns. This method immediately adds data samples as new nodes to existing networks. The probability of adding a new node is determined by the distance between the first, second, and third closest nodes. We applied our method for monitoring a moving object at its pace to demonstrate the usefulness of the proposed model. In addition, optimization methods are used such that processing can be performed in real-time.

Evaluating the relationship between two periodically correlated processes with Mandelbrot-Van Ness fractional Brownian motion errors using periodic copula

In many scientific fields such as meteorology, climatology and hydrology, one of major problems is to measure the relationships between some variables. To this aim, numerous parametric or non-parametric techniques have been introduced. But the applications of these approaches are usually restricted for some cases, for example, normal or stationary datasets. In this work, two periodically correlated time series with Mandelbrot-Van Ness fractional Brownian motion errors are considered and a novel technique to measure the relationship between them is introduced. The ability and power of the proposed method is evaluated using a deep simulation study. Moreover, the proposed procedure is used in a real world problem.

Deep Learning Based Prediction of Gas Chromatographic Retention Indices for a Wide Variety of Polar and Mid-Polar Liquid Stationary Phases.

Prediction of gas chromatographic retention indices based on compound structure is an important task for analytical chemistry. The predicted retention indices can be used as a reference in a mass spectrometry library search despite the fact that their accuracy is worse in comparison with the experimental reference ones. In the last few years, deep learning was applied for this task. The use of deep learning drastically improved the accuracy of retention index prediction for non-polar stationary phases. In this work, we demonstrate for the first time the use of deep learning for retention index prediction on polar (e.g., polyethylene glycol, DB-WAX) and mid-polar (e.g., DB-624, DB-210, DB-1701, OV-17) stationary phases. The achieved accuracy lies in the range of 16–50 in terms of the mean absolute error for several stationary phases and test data sets. We also demonstrate that our approach can be directly applied to the prediction of the second dimension retention times (GC × GC) if a large enough data set is available. The achieved accuracy is considerably better compared with the previous results obtained using linear quantitative structure-retention relationships and ACD ChromGenius software. The source code and pre-trained models are available online.

Improvement of the Nonparametric Estimation of Functional Stationary Time Series Using Yeo-Johnson Transformation with Application to Temperature Curves

In this article, Box-Cox and Yeo-Johnson transformation models are applied to two time series datasets of monthly temperature averages to improve the forecast ability. An application algorithm was proposed to transform the positive original responses using the first model and the stationary responses using the second model to improve the nonparametric estimation of the functional time series. The Box-Cox model contributed to improving the results of the nonparametric estimation of the original data, but the results become somewhat confusing after attempting to make the transformed response variable stationary in the mean, while the functional time series predictions were more accurate using the transformed stationary datasets using the Yeo-Johnson model.

Time series prediction using deep echo state networks

Artificial neural networks have been used for time series modeling and forecasting in many domains. However, they are often limited in their handling of nonlinear and chaotic data. More recently, reservoir-based recurrent neural net systems, most notably echo state networks (ESN), have made substantial improvements for time series modeling. Their shallow nature lends themselves to an efficient training method, but has limitations on nonstationary, nonlinear chaotic time series, particularly large, multidimensional time series. In this paper, we propose a novel approach for forecasting time series data based on an additive decomposition (AD) applied to the time series as a preprocessor to a deep echo state network. We compare the performance of our method, AD-DeepESN, on popular neural net architectures used for time series prediction. Stationary and nonstationary data sets are used to evaluate the performance of the methods. Our results are compelling, demonstrating that AD-DeepESN has superior performance, particularly on the most challenging time series that exhibit non-stationarity and chaotic behavior compared to existing methods.

Wavelet and multiple linear regression analysis for identifying factors affecting particulate matter PM2.5 in Mumbai City, India

Purpose The purpose of this paper is to identify significant factors of environmental variables and pollutants that have an effect on PM2.5 through wavelet and regression analysis. Design/methodology/approach In order to provide stable data set for regression analysis, multiresolution analysis using wavelets is conducted. For the sampled data, multicollinearity among the independent variables is removed by using principal component analysis and multiple linear regression analysis is conducted using PM2.5 as a dependent variable. Findings It is found that few pollutants such as NO2, NOx, SO2, benzene and environmental factors such as ambient temperature, solar radiation and wind direction affect PM2.5. The regression model developed has high R2 value of 91.9 percent, and the residues are stationary and not correlated indicating a sound model. Research limitations/implications The research provides a framework for extracting stationary data and other important features such as change points in mean and variance, using the sample data for regression analysis. The work needs to be extended across all areas in India and for various other stationary data sets there can be different factors affecting PM2.5. Practical implications Control measures such as control charts can be implemented for significant factors. Social implications Rules and regulations can be made more stringent on the factors. Originality/value The originality of this paper lies in the integration of wavelets with regression analysis for air pollution data.

Adaptive Online Learning With Regularized Kernel for One-Class Classification

In the past few years, kernel-based one-class extreme learning machine (ELM) receives quite a lot of attention by researchers for offline/batch learning due to its noniterative and fast learning capability. This paper extends this concept for adaptive online learning with regularized kernel-based one-class ELM classifiers for detection of outliers, and are collectively referred to as ORK-OCELM. Two frameworks, viz., boundary and reconstruction, are presented to detect the target class in ORK-OCELM. The kernel hyperplane-based baseline one-class ELM model considers whole data in a single chunk, however, the proposed one-class classifiers are adapted in an online fashion from the stream of training samples. The performance of ORK-OCELM is evaluated on a standard benchmark as well as synthetic datasets for both types of environments, i.e., stationary and nonstationary. While evaluating on stationary datasets, these classifiers are compared against batch learning-based one-class classifiers. Similarly, while evaluating on nonstationary datasets, the comparison is done with incremental learning-based online one-class classifiers. The results indicate that the proposed classifiers yield better or similar outcomes for both. In the nonstationary dataset evaluation, adaptability of the proposed classifiers in a changing environment is also demonstrated. It is further shown that the proposed classifiers have large stream data handling capability even under limited system memory. Moreover, the proposed classifiers gain significant time improvement compared to traditional online one-class classifiers (in all aspects of training and testing). A faster learning ability of the proposed classifiers makes them more suitable for real-time anomaly detection.

Tracking Sparse Linear Classifiers.

In this paper, we investigate the problem of sparse online linear classification in changing environments. We first analyze the tracking performance of standard online linear classifiers, which use gradient descent for minimizing the regularized hinge loss. The derived shifting bounds highlight the importance of choosing appropriate step sizes in the presence of concept drifts. Notably, we show that a better adaptability to concept drifts can be achieved using constant step sizes rather than the state-of-the-art decreasing step sizes. Based on these observations, we then propose a novel sparse approximated linear classifier, called sparse approximated linear classification (SALC), which uses a constant step size. In essence, SALC simply rounds small weights to zero for achieving sparsity and controls the truncation error in a principled way for achieving a low tracking regret. The degree of sparsity obtained by SALC is continuous and can be controlled by a parameter which captures the tradeoff between the sparsity of the model and the regret performance of the algorithm. Experiments on nine stationary data sets show that SALC is superior to the state-of-the-art sparse online learning algorithms, especially when the solution is required to be sparse; on seven groups of nonstationary data sets with various total shifting amounts, SALC also presents a good ability to track drifts. When wrapped with a drift detector, SALC achieves a remarkable tracking performance regardless of the total shifting amount.

Real time high cycle fatigue estimation algorithm and load history monitoring for vehicles by the use of frequency domain methods

Abstract Real time high cycle fatigue estimation problem for vehicles is examined by the use of frequency domain methods. The purpose was twofold: monitoring of fatigue damage and tracking of load history in real time. Firstly, power spectral density functions (PSDs) of acceleration measurement at a selected location are calculated in a piecewise manner by dividing the acceleration-time history into pieces. Following, Frequency Response Functions (FRF’s), whose outputs are the absolute maximum principal stress values at selected components, are calculated by finite element methods to account for multi-axial stress state in fatigue life estimations. Then, fatigue damage intensity at selected output locations is estimated using the FRF results. To this end, the following frequency domain fatigue estimation methods (FDFEMs) proposed for Gaussian and stationary data sets are applied to the selected components of a heavy duty truck: narrow-band approximation, Tovo and Benasciutti, Zhao and Baker, Dirlik and Tovo’s α 0.75 methods. Gaussianity and stationarity of acceleration-time data set used to estimate fatigue life of a component is checked to ensure the validity of FDFEMs. Numerical results are compared with experimental fatigue lives and damage calculations in time domain made by the combination of rainflow counting and Miner-Palmgren rules. There are two difficulties in implementing this approach using on-board equipment in real time such as overcoming the limited memory to store data sets and completing the computations sufficiently fast. To overcome them, the proposed approach is implemented in a piecewise manner and associated normalized PSDs are updated accordingly. Then, spectral moments and damage intensities are calculated in frequency domain. Implementation of the proposed approach is described in detail and numerical results are presented. It is shown that the proposed approach is able to predict the fatigue damage accurately and can keep track of loading conditions in real time.

Markerless positional verification using template matching and triangulation of kV images acquired during irradiation for lung tumors treated in breath-hold

Lung tumors treated in breath-hold are subject to inter- and intra-breath-hold variations, which makes tumor position monitoring during each breath-hold important. A markerless technique is desirable, but limited tumor visibility on kV images makes this challenging. We evaluated if template matching + triangulation of kV projection images acquired during breath-hold stereotactic treatments could determine 3D tumor position. Band-pass filtering and/or digital tomosynthesis (DTS) were used as image pre-filtering/enhancement techniques. On-board kV images continuously acquired during volumetric modulated arc irradiation of (i) a 3D-printed anthropomorphic thorax phantom with three lung tumors (n = 6 stationary datasets, n = 2 gradually moving), and (ii) four patients (13 datasets) were analyzed. 2D reference templates (filtered DRRs) were created from planning CT data. Normalized cross-correlation was used for 2D matching between templates and pre-filtered/enhanced kV images. For 3D verification, each registration was triangulated with multiple previous registrations. Generally applicable image processing/algorithm settings for lung tumors in breath-hold were identified. For the stationary phantom, the interquartile range of the 3D position vector was on average 0.25 mm for 12° DTS + band-pass filtering (average detected positions in 2D = 99.7%, 3D = 96.1%, and 3D excluding first 12° due to triangulation angle = 99.9%) compared to 0.81 mm for band-pass filtering only (55.8/52.9/55.0%). For the moving phantom, RMS errors for the lateral/longitudinal/vertical direction after 12° DTS + band-pass filtering were 1.5/0.4/1.1 mm and 2.2/0.3/3.2 mm. For the clinical data, 2D position was determined for at least 93% of each dataset and 3D position excluding first 12° for at least 82% of each dataset using 12° DTS + band-pass filtering. Template matching + triangulation using DTS + band-pass filtered images could accurately determine the position of stationary lung tumors. However, triangulation was less accurate/reliable for targets with continuous, gradual displacement in the lateral and vertical directions. This technique is therefore currently most suited to detect/monitor offsets occurring between initial setup and the start of treatment, inter-breath-hold variations, and tumors with predominantly longitudinal motion.

Detecting, anticipating, and predicting critical transitions in spatially extended systems.

A data-driven linear framework for detecting, anticipating, and predicting incipient bifurcations in spatially extended systems based on principal oscillation pattern (POP) analysis is discussed. The dynamics are assumed to be governed by a system of linear stochastic differential equations which is estimated from the data. The principal modes of the system together with corresponding decay or growth rates and oscillation frequencies are extracted as the eigenvectors and eigenvalues of the system matrix. The method can be applied to stationary datasets to identify the least stable modes and assess the proximity to instability; it can also be applied to nonstationary datasets using a sliding window approach to track the changing eigenvalues and eigenvectors of the system. As a further step, a genuinely nonstationary POP analysis is introduced. Here, the system matrix of the linear stochastic model is time-dependent, allowing for extrapolation and prediction of instabilities beyond the learning data window. The methods are demonstrated and explored using the one-dimensional Swift-Hohenberg equation as an example, focusing on the dynamics of stochastic fluctuations around the homogeneous stable state prior to the first bifurcation. The POP-based techniques are able to extract and track the least stable eigenvalues and eigenvectors of the system; the nonstationary POP analysis successfully predicts the timing of the first instability and the unstable mode well beyond the learning data window.

REVISITING EVIDENCE OF CHAOS IN X-RAY LIGHT CURVES: THE CASE OF GRS 1915+105

ABSTRACT Nonlinear time series analysis has been widely used to search for signatures of low-dimensional chaos in light curves emanating from astrophysical bodies. A particularly popular example is the microquasar GRS 1915+105, whose irregular but systematic X-ray variability has been well studied using data acquired by the Rossi X-ray Timing Explorer. With a view to building simpler models of X-ray variability, attempts have been made to classify the light curves of GRS 1915+105 as chaotic or stochastic. Contrary to some of the earlier suggestions, after careful analysis, we find no evidence for chaos or determinism in any of the GRS 1915+105 classes. The dearth of long and stationary data sets representing all the different variability classes of GRS 1915+105 makes it a poor candidate for analysis using nonlinear time series techniques. We conclude that either very exhaustive data analysis with sufficiently long and stationary light curves should be performed, keeping all the pitfalls of nonlinear time series analysis in mind, or alternative schemes of classifying the light curves should be adopted. The generic limitations of the techniques that we point out in the context of GRS 1915+105 affect all similar investigations of light curves from other astrophysical sources.

Technical note: Multiple wavelet coherence for untangling scale-specific and localized multivariate relationships in geosciences

Abstract. The scale-specific and localized bivariate relationships in geosciences can be revealed using bivariate wavelet coherence. The objective of this study was to develop a multiple wavelet coherence method for examining scale-specific and localized multivariate relationships. Stationary and non-stationary artificial data sets, generated with the response variable as the summation of five predictor variables (cosine waves) with different scales, were used to test the new method. Comparisons were also conducted using existing multivariate methods, including multiple spectral coherence and multivariate empirical mode decomposition (MEMD). Results show that multiple spectral coherence is unable to identify localized multivariate relationships, and underestimates the scale-specific multivariate relationships for non-stationary processes. The MEMD method was able to separate all variables into components at the same set of scales, revealing scale-specific relationships when combined with multiple correlation coefficients, but has the same weakness as multiple spectral coherence. However, multiple wavelet coherences are able to identify scale-specific and localized multivariate relationships, as they are close to 1 at multiple scales and locations corresponding to those of predictor variables. Therefore, multiple wavelet coherence outperforms other common multivariate methods. Multiple wavelet coherence was applied to a real data set and revealed the optimal combination of factors for explaining temporal variation of free water evaporation at the Changwu site in China at multiple scale-location domains. Matlab codes for multiple wavelet coherence were developed and are provided in the Supplement.

Flux–Variance Similarity in Complex Terrain and Its Sensitivity to Different Methods of Treating Non-stationarity

Various criteria have been developed to remove non-stationarity in turbulence time series, though it remains unclear how the choice of the stationarity criterion affects similarity functions in the framework of the Monin–Obukhov similarity theory. To investigate this, we use stationary datasets that result from applying five common criteria to remove non-stationarity in turbulence time series from the Terrain-Induced Rotor EXperiment conducted in Owens Valley, California. We determine the form of the flux–variance similarity functions and the scatter around these similarity functions for all five stationary datasets. Data were collected at two valley locations and one slope location using 34-m flux towers with six levels of turbulence measurements. Our results show (i) systematic differences from previously found near-neutral values of the parameters in the flux–variance similarity functions over flat terrain, indicating a larger anisotropy of the flow over complex than over flat terrain, (ii) a reduction of this anisotropy when stationary data are used, with the amount of reduction depending on the stationarity criterion, (iii) a general reduction in scatter around the similarity functions when using stationary data but more so for stable than for unstable stratification, and for valley locations than for the slope location, and (iv) a weak variation with height of near-neutral values of parameters in the flux–variance similarity functions.

Validation of reference genes for quantitative real-time PCR (qPCR) analysis of Actinobacillus suis.

BackgroundQuantitative real-time PCR is a valuable tool for evaluating bacterial gene expression. However, in order to make best use of this method, endogenous reference genes for expression data normalisation must first be identified by carefully validating the stability of expression under experimental conditions. Therefore, the objective of this study was to validate eight reference genes of the opportunistic swine pathogen, Actinobacillus suis, grown in aerobic cultures with (Epinephrine) or without (Aerobic) epinephrine in the growth medium and in anoxic static cultures (Anoxic), and sampled during exponential and stationary phases.ResultsUsing the RefFinder tool, expression data were analysed to determine whether comprehensive stability rankings of selected reference genes varied with experimental design. When comparing Aerobic and Epinephrine cultures by growth phase, pyk and rpoB were both among the most stably expressed genes, but when analysing both growth phases together, only pyk remained in the top three rankings. When comparing Aerobic and Anoxic samples, proS ranked among the most stable genes in exponential and stationary phase data sets as well as in combined rankings. When analysing the Aerobic, Epinephrine, and Anoxic samples together, only gyrA ranked consistently among the top three most stably expressed genes during exponential and stationary growth as well as in combined rankings; the rho gene ranked as least stably expressed gene in this data set.ConclusionsReference gene stability should be carefully assessed with the design of the experiment in mind. In this study, even the commonly used reference gene 16S rRNA demonstrated large variability in stability depending on the conditions studied and how the data were analysed. As previously suggested, the best approach may be to use a geometric mean of multiple genes to normalise qPCR results. As researchers continue to validate reference genes for various organisms in multiple growth conditions and sampling time points, it may be possible to make informed predictions as to which genes may be most suitable to validate for a given experimental design, but in the meantime, the reference genes used to normalise qPCR data should be selected with caution.

Application of Hybrid Techniques (Self-Organizing Map and Fuzzy Algorithm) Using Backscatter Data for Segmentation and Fine-Scale Roughness Characterization of Seepage-Related Seafloor Along the Western Continental Margin of India

<?Pub Dtl=""?> In this paper, a seafloor characterization technique to unravel the number of data classes using multibeam echo-sounding backscatter data has been demonstrated. An application of self-organizing maps (SOMs) to backscatter profile data has been developed to determine the number of classes. Consequently, the fuzzy C-means (FCM) method is employed thereafter, using the number of class information of the backscatter profiles segmentation. The use of the soft-computational technique for seafloor backscatter data facilitates in achieving stationary profile data sets suitable for seafloor roughness model application. The power spectral density (PSD) function of the segmented profiles provides the power law parameters ( <formula formulatype="inline" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex Notation="TeX">$\beta$</tex> </formula> and <formula formulatype="inline" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex Notation="TeX">$a'$</tex> </formula> ) through curve fitting, using power law expression. The data acquired from western continental margin of India (WCMI), off Goa, reveal five distinct classes of backscatter strength having different segment lengths. The estimated roughness parameters ( <formula formulatype="inline" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex Notation="TeX">$\beta$</tex> </formula> and <formula formulatype="inline" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex Notation="TeX">$a'$</tex> </formula> ) of the segmented profiles provide quantitative information about the area seafloor roughness. A gridded map of the estimated roughness parameter <formula formulatype="inline" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex Notation="TeX">$(\beta)$</tex></formula> is generated using the “krigging” method. The gridded map and the class of the segmented profiles overlaid on the backscatter map have been presented, which is a first time application relevant for the understanding of the seafloor. The present study underscores a combination of soft-computational (SOM and FCM) and numerical techniques (power spectral density at short and fine scales) to effectively recognize the seafloor processes and the associated sedimentological dynamics in a complex geographical environment (including the pockmarks and faulted structures), which is subjected to strong bottom currents and seasonal upwelling.