Continuous Dataset Research Articles

Benefits of the Open-Loop Tracking Command (OLTC): Extending conventional nadir altimetry to inland waters monitoring

Abstract In the past 25 years, radar altimeters have been designed and operated primarily for ocean observation. However, over the past decade there has been a growing interest for altimetry measurements over inland waters. Studying lakes, reservoirs and rivers water level is of prime importance for the hydrology community to assess the Earth’s global resources of fresh water. Satellite altimetry is key to providing such global and continuous datasets of water surface height. Indeed, much progress has been made in altimeters capability to acquire quality measurements over inland waters. In this paper we present an overview of major technical evolutions of the tracking function of altimeters, from Jason-2/POS3 and SARAL/AltiKa to Jason-3/POS3B and Sentinel-3/SRAL and the improvements brought by the Open-Loop Tracking Command (OLTC) to extend the altimeter observation domain to inland waters. The OLTC is an on-board feature used to set the altimeter waveforms reception window, containing a priori elevation information built from a dedicated database of hydrological targets. This paper focuses on the use of OLTC for acquiring reliable water signal rather than discussing the accuracy of the retrieved height, which constitutes a further step in the altimeter data processing chain. We show how Jason-3 and Sentinel-3 altimetry missions are currently able to observe and monitor about 70,000 hydrological targets over the globe and how it is contributing in building a global dataset of inland waters level, in preparation for future missions such as Sentinel-6 and the Surface Water Ocean Topography mission (SWOT).

A Novel Sign Language Recognition Framework Using Hierarchical Grassmann Covariance Matrix

Visual sign language recognition is an interesting and challenging problem. To create a discriminative representation, a hierarchical Grassmann covariance matrix (HGCM) model is proposed for sign description. Furthermore, a multi-temporal belief propagation (MTBP) based segmentation approach is presented for continuous sequence spotting. Concretely speaking, a sign is represented by multiple covariance matrices, followed by evaluating and selecting their most significant singular vectors. These covariance matrices are transformed into a more compact and discriminative HGCM, which is formulated on the Grassmann manifold. Continuous sign sequences can be recognized frame by frame using the HGCM model, before being optimized by MTBP, which is a carefully designed graphic model. The proposed method is thoroughly evaluated on isolated and synthetic and real continuous sign datasets as well as on HDM05. Extensive experimental results convincingly show the effectiveness of our proposed framework.

The Main Factor Determining the Dynamics of the Lake Ecosystem under Excessive Nutrient Loading (A Case Study of the Naroch Lakes)

The system of the Naroch lakes, which includes the eutrophic Lake Batorino, the mesotrophic Lake Myastro, and the oligo-mesotrophic Lake Naroch, serves as a model object for the study of the factors initially influencing the state of a particular lake ecosystem affected by variable nutrient loading. Throughout the 1970s–2000s, these lakes have been going through the stages of anthropogenic eutrophication, deeutrophication, and benthification. A set of continuous data based on the seasonal means ​​of the eight parameters from each lake for the period 1978–2015 has been analyzed by Principal Component and Singular Spectrum Analysis (SSA, or Caterpillar). In addition, we consider the dynamics of the trophic index of each lake, which had been calculated earlier at the same data set. The first main component is the stability of the lake ecosystem under variable nutrient loading, which is inversely related to the trophic state of the water body. This component determines the condition of the Batorino, Myastro and Naroch lakes by 63, 65, and 43% respectively.

A spatio-temporal continuous soil moisture dataset over the Tibet Plateau from 2002 to 2015

Surface soil moisture is a key variable in the exchange of water and energy between the land surface and the atmosphere, and critical to meteorology, hydrology, and ecology. The Tibetan Plateau (TP), known as “The third pole of the world” and “Asia’s water towers”, exerts huge influences on and sensitive to global climates. In this situation, longer time series of soil moisture can provide sufficient information to understand the role of the TP. This paper presents the first comprehensive dataset (2002–2015) of spatio-temporal continuous soil moisture at 0.25° resolution, based on satellite-based optical (i.e. MODIS) and microwave (ECV) products using a machine learning method named general regression neural network (GRNN). The dataset itself reveals significant information on the soil moisture and its changes over the TP, and can aid to understand the potential driven mechanisms for climate change over the TP.

Analysis of Low-level Temperature Inversions and Their Effects on Aerosols in the Lower Atmosphere

High-quality and continuous radiosonde, aerosol and surface meteorology datasets are used to investigate the statistical characteristics of meteorological parameters and their effects on aerosols. The data were collected at the Atmospheric Radiation Measurement Southern Great Plains climate research facility during 2000–15. The parameters and vertical distribution of temperature inversion layers were found to have strong diurnal and seasonal changes. For surface-based temperature inversion (SBI), the mean frequency and depth of temperature inversion layers were 39.4% and 198 m, respectively. The temperature difference between the top and bottom of SBI was 4.8°C, and so the temperature gradient was 2.4°C (100 m)−1. The detailed vertical distributions of temperature inversion had been determined, and only the temperature inversion layers below 1000 m showed diurnal and seasonal variations. Mean surface aerosol number concentrations increased by 43.0%, 21.9% and 49.2% when SBIs were present at 0530, 1730 and 2330 LST, respectively. The effect of SBI on surface aerosol concentration was weakest in summer (18.1%) and strongest in winter (58.4%). During elevated temperature inversion events, there was no noticeable difference in surface aerosol number concentrations. Temperature differences and temperature gradients across SBIs correlated fairly well with aerosol number concentrations, especially for temperature gradients. The vertical distribution of aerosol optical properties with and without temperature inversions was different. Surface aerosol measurements were representative of the air within (below), but not above, SBIs and EIs. These results provide a basis for developing a boundary layer aerosol accumulation model and for improving radiative transfer models in the lower atmosphere.

Alternative models for the modified form of ridge regularized linear model in discovering Markov boundary

It has been proved that the modified form of ridge regularized linear model (MRRLM) can discover a subset of Markov boundary of the target variable under some constrained conditions. However, MRRLM cannot be applied to the data sets with collinear variables due to covariance matrix is employed. To develop a suitable alternative model for MRRLM, we study the relationships of discovery performance of Markov boundary among MRRLM, ridge regression linear model (RRLM), and LASSO combining with permutation test through empirical method. In addition, we also proposed a new NVRRLM to discover Markov boundary of the target variable. The experimental results show that: (1) On the binary data sets, MRRLM has a basically similar performance with LASSO and RRLM; (2) On the continuous data sets, MRRLM has a basically similar discovery performance with LASSO but has higher discovery performance than RRLM; (3) The new NVRRLM can replace MRRLM on the data sets with collinear variables. The above experimental results demonstrate NVRRLM can effectively deal with variable collinearity problems.

Comparison and Assessment of Regional and Global Land Cover Datasets for Use in CLASS over Canada

Global land cover information is required to initialize land surface and Earth system models. In recent years, new land cover (LC) datasets at finer spatial resolutions have become available while those currently implemented in most models are outdated. This study assesses the applicability of the Climate Change Initiative (CCI) LC product for use in the Canadian Land Surface Scheme (CLASS) through comparison with finer resolution datasets over Canada, assisted with reference sample data and a vegetation continuous field tree cover fraction dataset. The results show that in comparison with the finer resolution maps over Canada, the 300 m CCI product provides much improved LC distribution over that from the 1 km GLC2000 dataset currently used to provide initial surface conditions in CLASS. However, the CCI dataset appears to overestimate needleleaf forest cover especially in the taiga-tundra transition zone of northwestern Canada. This may have partly resulted from limited availability of clear sky MEdium Resolution Imaging Spectrometer (MERIS) images used to generate the CCI classification maps due to the long snow cover season in Canada. In addition, changes based on the CCI time series are not always consistent with those from the MODIS or a Landsat-based forest cover change dataset, especially prior to 2003 when only coarse spatial resolution satellite data were available for change detection in the CCI product. It will be helpful for application in global simulations to determine whether these results also apply to other regions with similar landscapes, such as Eurasia. Nevertheless, the detailed LC classes and finer spatial resolution in the CCI dataset provide an improved reference map for use in land surface models in Canada. The results also suggest that uncertainties in the current cross-walking tables are a major source of the often large differences in the plant functional types (PFT) maps, and should be an area of focus in future work.

Assessment of regional aerosol radiative effects under the SWAAMI campaign – Part 1: Quality-enhanced estimation of columnar aerosol extinction and absorption over the Indian subcontinent

Abstract. Improving the accuracy of regional aerosol climate impact assessment calls for improvement in the accuracy of regional aerosol radiative effect (ARE) estimation. One of the most important means of achieving this is to use spatially homogeneous and temporally continuous datasets of critical aerosol properties, such as spectral aerosol optical depth (AOD) and single scattering albedo (SSA), which are the most important parameters for estimating aerosol radiative effects. However, observations do not provide the above; the space-borne observations though provide wide spatial coverage, are temporal snapshots and suffer from possible sensor degradation over extended periods. On the other hand, the ground-based measurements provide more accurate and temporally continuous data but are spatially near-point observations. Realizing the need for spatially homogeneous and temporally continuous datasets on one hand and the near non-existence of such data over the south Asian region (which is one of the regions where aerosols show large heterogeneity in most of their properties), construction of accurate gridded aerosol products by synthesizing the long-term space-borne and ground-based data has been taken up as an important objective of the South West Asian Aerosol Monsoon Interactions (SWAAMI), a joint Indo-UK field campaign, aiming at characterizing aerosol–monsoon links and their variabilities over the Indian region. In Part 1 of this two-part paper, we present spatially homogeneous gridded datasets of AOD and absorption aerosol optical depth (AAOD), generated for the first time over this region. These data products are developed by merging the highly accurate aerosol measurements from the dense networks of 44 (for AOD) and 34 (for AAOD) ground-based observatories of Aerosol Radiative Forcing over India NETwork (ARFINET) and AErosol RObotic NETwork (AERONET) spread across the Indian region, with satellite-retrieved AOD and AAOD, following statistical assimilation schemes. The satellite data used for AOD assimilation include AODs retrieved from MODerate Imaging Spectroradiometer (MODIS) and Multiangle Imaging SpectroRadiometer (MISR) over the same domain. For AAOD, the ground-based black carbon (BC) mass concentration measurements from the network of 34 ARFINET observatories and satellite-based (Kalpana-1, INSAT-3A) infrared (IR) radiance measurements are blended with gridded AAODs (500 nm, monthly mean) derived from Ozone Monitoring Instrument (OMI)-retrieved AAODs (at 354 and 388 nm). The details of the assimilation methods and the gridded datasets generated are presented in this paper. The merged gridded AOD and AAOD products thus generated are validated against the data from independent ground-based observatories, which were not used for the assimilation process but are representative of different subregions of the complex domain. This validation exercise revealed that the independent ground-based measurements are better confirmed by merged datasets than the respective satellite products. As ensured by assimilation techniques employed, the uncertainties in merged AODs and AAODs are significantly less than those in corresponding satellite products. These merged products also all exhibit important large-scale spatial and temporal features which are already reported for this region. Nonetheless, the merged AODs and AAODs are significantly different in magnitude from the respective satellite products. On the background of above-mentioned quality enhancements demonstrated by merged products, we have employed them for deriving the columnar SSA and analysed its spatiotemporal characteristics. The columnar SSA thus derived has demonstrated distinct seasonal variation over various representative subregions of the study domain. The uncertainties in the derived SSA are observed to be substantially less than those in OMI SSA. On the backdrop of these benefits, the merged datasets are employed for the estimation of regional aerosol radiative effects (direct), the results of which would be presented in a companion paper, Part 2 of this two-part paper.

Deriving a long-term pan evaporation reanalysis dataset for two Chinese pan types

A long-term continuous and consistent pan evaporation (Epan) reanalysis dataset will augment the analysis of Epan distributions when the observation network is discontinuous or inconsistent, and assist in the evaluation of the outputs of General Circulation Models (GCMs) and Land Surface Models (LSMs). From the 1950s to early 2000s, China had a continuous observation of the D20 pan, but this was replaced by the 601B pan across China around 2002, and thus the Epan observation network became discontinuous and inconsistent. This study developed a long-term monthly, 0.05°, continuous and consistent reanalysis dataset for both D20 and 601B pans covering mainland China throughout 1960–2014, based on meteorological data homogenization and interpolation and Epan assimilation. The PenPan-V3 model used in Epan assimilation was successfully validated by observations at 767 and 591 stations for D20 and 601B pans, respectively. Comprehensively considering the physical influence of elevation, radiation, wind speed, humidity, and air temperature, the average annual and seasonal gridded Epan reanalyses show significant spatial dependent on proximity to the ocean and latitude, consistent with previous studies. The reanalysis dataset can be used to analyze Epan distributions across China, including the areas without observations, and to estimate the representativeness of Epan to atmospheric evaporative demand. The dataset has been released in two cloud servers in China and the United States, and it will continue to be maintained and updated.

Bat algorithm and k-means techniques for classification performance improvement

This paper presents Bat Algorithm and K-Means techniques for classification performance improvement. The objective of this study is to investigate efficiency of Bat Algorithm in discrete dataset and to find the optimum feature in discrete dataset. In this study, one technique that comprise the discretization technique and feature selection technique have been proposed. Our contribution is in two process of classification: pre-processing and feature selection process. First, to proposed discretization techniques called as BkMD, where we hybrid Bat Algorithm technique and K-Means classifier. Second, to proposed BkMDFS as feature selection technique where Bat Algorithm is embed into BkMD. In order to evaluate our proposed techniques, 14 continuous dataset from various applications are used in experiment. From the experiment, results show that BkMDFS outperforms in most performance measures. Hence it shows that, Bat Algorithm have potential to be one of the discretization technique and feature selection technique.

The Consistency of SSM/I vs. SSMIS and the Influence on Snow Cover Detection and Snow Depth Estimation over China

The long-term variations in snow depth are important in hydrological, meteorological, and ecological implications and climatological studies. The series of Special Sensor Microwave/Imager (SSM/I) and Special Sensor Microwave Imager Sounder (SSMIS) instruments onboard the Defense Meteorological Satellite Program (DMSP) platforms has provided a consistent 30+ year data record of global observations that is well-suited for the estimation of snow cover, snow depth, and snow water equivalent (SWE). To maximize the use of this continuous microwave observation dataset in long-term snow analysis and obtain an objective result, consistency among the SSM/I and SSMIS sensors is required. In this paper, we evaluated the consistency between the SSM/I and SSMIS concerning the observed brightness temperature (Tb) and the retrieved snow cover area and snow depth from January 2007 to December 2008, where the F13 SSM/I and the F17 SSMIS overlapped. Results showed that Tb bias at 19 GHz spans from −2 to −3 K in snow winter seasons, and from −4 to −5 K in non-snow seasons. There is a slight Tb bias at 37 GHz from −2 to 2 K, regardless of season. For 85 (91) GHz, the bias presents some uncertainty from the scattering effect of the snowpack and atmospheric emission. The overall consistency between SSM/I and SSMIS with respect to snow cover detection is between 80% and 100%, which will result in a maximum snow cover area difference of 25 × 104 km2 in China. The inconsistency in Tb between SSM/I and SSMIS can result in a −2 and −0.67 cm snow depth bias for the dual-channel and multichannel algorithms, respectively. SSMIS tends to yield lower snow depth estimates than SSM/I. Moreover, there are notable bias differences between SSM/I- and SSMIS-estimated snow depths in the tundra and taiga snow classes. Our results indicate the importance of considering the Tb bias in microwave snow cover detection and snow depth retrieval and point out that, due to the sensitivity of bias to seasons, it is better to do the intercalibration with a focus on snow-covered winter seasons. Otherwise, the bias in summer will disturb the calibration coefficients and introduce more error into the snow retrievals if the seasonal difference is not carefully evaluated and separated.

Investigation of Arctic middle-atmospheric dynamics using 3 years of H<sub>2</sub>O and O<sub>3</sub> measurements from microwave radiometers at Ny-Ålesund

Abstract. We used 3 years of water vapour and ozone measurements to study the dynamics in the Arctic middle atmosphere. We investigated the descent of water vapour within the polar vortex, major and minor sudden stratospheric warmings and periodicities at Ny-Ålesund. The measurements were performed with the two ground-based microwave radiometers MIAWARA-C and GROMOS-C, which have been co-located at the AWIPEV research base at Ny-Ålesund, Svalbard (79∘ N, 12∘ E), since September 2015. Both instruments belong to the Network for the Detection of Atmospheric Composition Change (NDACC). The almost continuous datasets of water vapour and ozone are characterized by a high time resolution of the order of hours. A thorough intercomparison of these datasets with models and measurements from satellite, ground-based and in situ instruments was performed. In the upper stratosphere and lower mesosphere the MIAWARA-C water vapour profiles agree within 5 % with SD-WACCM simulations and ACE-FTS measurements on average, whereas AuraMLS measurements show an average offset of 10 %–15 % depending on altitude but constant in time. Stratospheric GROMOS-C ozone profiles are on average within 6 % of the SD-WACCM model, the AuraMLS and ACE-FTS satellite instruments and the OZORAM ground-based microwave radiometer which is also located at Ny-Ålesund. During these first 3 years of the measurement campaign typical phenomena of the Arctic middle atmosphere took place, and we analysed their signatures in the water vapour and ozone measurements. Two major sudden stratospheric warmings (SSWs) took place in March 2016 and February 2018 and three minor warmings were observed in early 2017. Ozone-rich air was brought to the pole and during the major warmings ozone enhancements of up to 4 ppm were observed. The reversals of the zonal wind accompanying a major SSW were captured in the GROMOS-C wind profiles which are retrieved from the ozone spectra. After the SSW in February 2018 the polar vortex re-established and the water vapour descent rate in the mesosphere was 355 m d−1. Inside of the polar vortex in autumn we found the descent rate of mesospheric water vapour from MIAWARA-C to be 435 m d−1 on average. We find that the water vapour descent rate from SD-WACCM and the vertical velocity w‾* of the residual mean meridional circulation from SD-WACCM are substantially higher than the descent rates of MIAWARA-C. w‾* and the zonal mean water vapour descent rate from SD-WACCM agree within 10 % after the SSW, whereas in autumn w‾* is up to 40 % higher. We further present an overview of the periodicities in the water vapour and ozone measurements and analysed seasonal and interannual differences.

The Use of Sentinel-1 Synthetic Aperture Radar (SAR) Images and Open-Source Software for Cultural Heritage: An Example from Paphos Area in Cyprus for Mapping Landscape Changes after a 5.6 Magnitude Earthquake

Active satellite remote sensors have emerged in the last years in the field of archaeology, providing new tools for monitoring extensive cultural heritage landscapes and areas. These active sensors, namely synthetic aperture radar (SAR) satellites, provide systematic datasets for mapping land movements triggered from earthquakes, landslides, and so on. Copernicus, the European program for monitoring the environment, provides continuous radar datasets through the Sentinel-1 mission with an almost worldwide coverage. This paper aims to demonstrate how the use of open-access and freely distributed datasets such as those under the Copernicus umbrella, along with the exploitation of open-source radar processing software, namely the sentinel applications platform (SNAP) and SNAPHU tools, provided respectively by the European Space Agency (ESA) and the University of Stanford, can be used to extract an SAR interferogram in the wider area of Paphos, located in the western part of Cyprus. The city includes various heritage sites and monuments, some of them already included in the UNESCO World Heritage list. The interferogram was prepared to study the effects of an earthquake to the buildings and sites of the area. The earthquake of a 5.6 magnitude on the Richter scale was triggered on 15 April 2015 and was strongly felt throughout the whole island. The interferogram results were based on Differential Synthetic Aperture Radar Interferometry (D-InSAR) methodology, finding a maximum uplift of 74 mm and a maximum subsidence of 31 mm. The overall process and methodology are presented in this paper.

A novel hybrid GA–PSO framework for mining quantitative association rules

Discovering association rules is a useful and common technique for data mining in which dependencies among datasets are shown. Discovering the rules from continuous numeric datasets is one of the common challenges in data mining. Furthermore, another restriction imposed by algorithms in this area is the need to determine the minimum threshold for the criteria of support and confidence. By drawing on two heuristic optimization techniques, to wit, the genetic algorithm (GA) and particle swarm optimization (PSO) algorithm, a hybrid algorithm for extracting quantitative association rules was developed in this research. Accurate and interpretable rules result from the integration of the multiple objectives GA with the multiple objective PSO algorithms, which redresses the balance in the exploitation and exploration tasks. The useful and appropriate rules and the most suitable numerical intervals are discovered by proposing a multi-criteria method in which there is no need to discretize numerical values and to determine threshold values of minimum support and confidence. Different criteria are used to determine appropriate rules. In this algorithm, the selected rules are extracted based on confidence, interestingness and comprehensibility. The results gained over five real-world datasets evidence the effectiveness of the proposed method. By hybridization of the GA and the PSO algorithm, the proposed approach has achieved considerable improvements compared with the basic algorithms in the criteria of the number of extracted rules from dataset, high confidence measure and support percentage.

Fuzzy kernel feature selection with multi-objective differential evolution algorithm

In this paper, we propose a multi-objective differential evolution-based filter approach for feature selection that interconnects fuzzy- and kernel-based information theory measures to find feature subsets that are optimal responses to the targets. In contrast to the existing filter approaches using the principles of information theory and rough set theory, our approach can be applied to continuous datasets without discretisation. Moreover, our study is the first in the literature that employs fuzzy and kernel measures to form a filter criterion for feature selection, to our knowledge. We prove various favourable results using a variety of benchmark datasets and also demonstrate that our approach can better search the dimensionality space to reach maximum predictive of the response.

The optimal upper bound of the number of queries for Laplace mechanism under differential privacy

Differential privacy is a state-of-the-art technology for privacy preserving, and Laplace mechanism is a simple and powerful tool to realize differential privacy. However, there is an obvious flaw in differential privacy, which is each query function can only be executed finite times for the reason that adversary can recover the real query result if he executes the same query function many times. Unfortunately, how to set the upper bound for the number of linear queries is still an issue. In this paper, we focus on the linear query function in Laplace-based mechanisms, and we propose a method to set the upper bound for the number of linear queries from the perspective of information theory. The main idea is, firstly we find the most aggressive linear query that leaks the maximum information about the dataset to adversaries, and we set the upper bound of the number of queries so that even if the most aggressive linear query cannot leak the whole self-information about any individual to the adversary. On the other hand, the number of queries is also influenced by the type of dataset (continuous and discrete). In this paper, we also discuss the different upper bound for the number of queries for continuous datasets and discrete datasets. Finally, we conduct the experiments on the continuous dataset and the discrete dataset to prove our result.

Unsupervised Large‐Scale Search for Similar Earthquake Signals

AbstractSeismology has continuously recorded ground‐motion spanning up to decades. Blind, uninformed search for similar‐signal waveforms within this continuous data can detect small earthquakes missing from earthquake catalogs, yet doing so with naive approaches is computationally infeasible. We present results from an improved version of the Fingerprint And Similarity Thresholding (FAST) algorithm, an unsupervised data‐mining approach to earthquake detection, now available as open‐source software. We use FAST to search for small earthquakes in 6–11 yr of continuous data from 27 channels over an 11‐station local seismic network near the Diablo Canyon nuclear power plant in central California. FAST detected 4554 earthquakes in this data set, with a 7.5% false detection rate: 4134 of the detected events were previously cataloged earthquakes located across California, and 420 were new local earthquake detections with magnitudes −0.3≤ML≤2.4, of which 224 events were located near the seismic network. Although seismicity rates are low, this study confirms that nearby faults are active. This example shows how seismology can leverage recent advances in data‐mining algorithms, along with improved computing power, to extract useful additional earthquake information from long‐duration continuous data sets.

Urban landslide monitoring by combined use of multiple methodologies - a case study on Sv. Anton town, Slovakia

ABSTRACTWe employed integrated methods to assess the landslide movement in Sv. Anton town in the Western Carpathians Neogene Volcanic Field (Central Slovakia). The integrated diagnostics required study of the landslide kinematic activity by a combination of Global Navigation Satellite Systems (GNSS) and Electrical Resistivity Tomography (ERT) imaging from November 2013 to March 2015. A topographic model with 2-cm accuracy was constructed from Unmanned Aerial Vehicles (UAV) photogrammetry. Continuous spatial datasets of movement and displacement field vectors were interpolated from the measured movements over the entire study period. Although deformation studies in Slovakia have a long-term tradition, complex interdisciplinary studies in urbanized areas are still lacking. This inspired our main objectives: to identify landslide kinematics and to reconstruct and define the rates of annual landslide movement obtained from geodetic measurement at the monitoring points. Our results demonstrate how landslide integrated diagnostics contribute to the detection of slope instability, with a maximum velocity of 60.82 mm/yr during the summer period. The precipitation effects are consistent with the Sv. Anton landslide displacement acceleration, and the following increases in total monthly precipitations are staggering compared to long-term monthly averages: July precipitation increased by 175.3%, August by 203.3%, and September by 198.1%.

Globally-Consistent Rule-Based Summary-Explanations for Machine Learning Models: Application to Credit-Risk Evaluation

We develop a method for interpreting specific predictions made by (global) predictive models by constructing (local) models tailored to each specific observation (these are also called explanations in the literature). Unlike existing work that explains'' specific observations by approximating global models in the vicinity of these observations, we fit models that are globally consistent with predictions made by the global model on past data. We focus on rule-based models (also known as association rules or conjunctions of predicates), which are interpretable and widely used in practice. We design multiple algorithms to extract such rules from discrete and continuous datasets, and study their theoretical properties. Finally, we apply these algorithms to multiple credit-risk models trained on real-world data from FICO and demonstrate that our approach effectively produces sparse summary-explanations of these models in short period of time. Our approach is model-agnostic (that is, can be used to interpret any predictive model), and solves a minimum set cover problem to construct its summaries.

The effect of running speed on joint coupling coordination and its variability in recreational runners

The purpose of this study was to examine the effect of speed on coordination and its variability in running gait using vector coding analysis. Lower extremity kinematic data were collected for thirteen recreational runners while running at three different speeds in random order: preferred speed, 15% faster and 15% lower than preferred speed. A dynamical systems approach, using vector coding and circular statistics, were used to quantify coordination and its variability for selected hip-knee and knee-ankle joint couplings. The influence of running speed was calculated from the continuous data sets of the running cycle, allowing for the identification of time percentages where differences existed. Results indicate that increases in running speed produced moderate alterations in the frequency of movement patterns which were not enough to alter classification of coordination. No effects of speed on coordination variability were observed. This study has demonstrated that coordination and coordination variability is generally stable in the range of ±15% around of preferred speed in recreational runners.