Use Of Limited Data Sets Research Articles

Enhanced detection of diabetes mellitus using novel ensemble feature engineering approach and machine learning model

Diabetes is a persistent health condition led by insufficient use or inappropriate use of insulin in the body. If left undetected, it can lead to further complications involving organ damage such as heart, lungs, and eyes. Timely detection of diabetes helps obtain the right medication, diet, and exercise plan to lead a healthy life. ML approach has been utilized to obtain rapid and reliable diabetes detection, however, existing approaches suffer from the use of limited datasets, lack of generalizability, and lower accuracy. This study proposes a novel feature extraction approach to overcome these limitations by using an ensemble of convolutional neural network (CNN) and long short-term memory (LSTM) models. Multiple datasets are combined to make a larger dataset for experiments and multiple features are utilized for investigating the efficacy of the proposed approach. Features from the extra tree classifier, CNN, and LSTM are also considered for comparison. Experimental results reveal the superb performance of CNN-LSTM-based features with random forest model obtaining a 0.99 accuracy score. This performance is further validated by comparison with existing approaches and k-fold cross-validation which shows the proposed approach provides robust results.

Use of an environmental proxy to determine turbulence regime surrounding a full-scale tidal turbine deployed within the Fromveur Strait, Brittany, France

Establishing a relationship between tidal current conditions and tidal turbine performance and loads is a critically important consideration for turbine reliability. Nonetheless, obtaining in-situ information is often challenging, and as a result both environmental and load data may be more sparse than desired. This study presents a method to make use of limited data sets by establishing a relationship between measurements of hydrodynamic variability and turbine power or blade strain variability, even when these measurements are not taken simultaneously. The method is tested on data from the deployment of a full-scale pilot tidal turbine: in situ velocity measurements and turbulence characteristics taken at times when the turbine was not installed were associated with power and strain measurements during the turbine’s deployment via a Delft3D proxy. The data show that the variability of active power correlates well with larger turbulence kinetic energy (TKE) when comparing similar populations via the proxy. Examination of blade strain variance against TKE shows a weaker correlation, with fat-tailed distributions and extremely high strain values prominent across all flow speeds. Acceleration or deceleration of the flow influenced the power variability of the turbine, with larger standard deviations recorded across accelerating flows. No significant difference was found when comparing blade strain variance in accelerating and decelerating flows. We conclude that the proxy method studied can establish a population-level relationship between non-simultaneous environmental and load data, but that the accuracy and precision of this relationship depends on the amount of data available: this method is therefore only suitable where there is a sufficiently rich dataset.

SAResNet: self-attention residual network for predicting DNA-protein binding.

Knowledge of the specificity of DNA-protein binding is crucial for understanding the mechanisms of gene expression, regulation and gene therapy. In recent years, deep-learning-based methods for predicting DNA-protein binding from sequence data have achieved significant success. Nevertheless, the current state-of-the-art computational methods have some drawbacks associated with the use of limited datasets with insufficient experimental data. To address this, we propose a novel transfer learning-based method, termed SAResNet, which combines the self-attention mechanism and residual network structure. More specifically, the attention-driven module captures the position information of the sequence, while the residual network structure guarantees that the high-level features of the binding site can be extracted. Meanwhile, the pre-training strategy used by SAResNet improves the learning ability of the network and accelerates the convergence speed of the network during transfer learning. The performance of SAResNet is extensively tested on 690 datasets from the ChIP-seq experiments with an average AUC of 92.0%, which is 4.4% higher than that of the best state-of-the-art method currently available. When tested on smaller datasets, the predictive performance is more clearly improved. Overall, we demonstrate that the superior performance of DNA-protein binding prediction on DNA sequences can be achieved by combining the attention mechanism and residual structure, and a novel pipeline is accordingly developed. The proposed methodology is generally applicable and can be used to address any other sequence classification problems.

Physics-guided convolutional neural network (PhyCNN) for data-driven seismic response modeling

Accurate prediction of building’s response subjected to earthquakes makes possible to evaluate building performance. To this end, we leverage the recent advances in deep learning and develop a physics-guided convolutional neural network (PhyCNN) for data-driven structural seismic response modeling. The concept is to train a deep PhyCNN model based on limited seismic input–output datasets (e.g., from simulation or sensing) and physics constraints, and thus establish a surrogate model for structural response prediction. Available physics (e.g., the law of dynamics) can provide constraints to the network outputs, alleviate overfitting issues, reduce the need of big training datasets, and thus improve the robustness of the trained model for more reliable prediction. The surrogate model is then utilized for fragility analysis given certain limit state criteria. In addition, an unsupervised learning algorithm based on K-means clustering is also proposed to partition the datasets to training, validation and prediction categories, so as to maximize the use of limited datasets. The performance of PhyCNN is demonstrated through both numerical and experimental examples. Convincing results illustrate that PhyCNN is capable of accurately predicting building’s seismic response in a data-driven fashion without the need of a physics-based analytical/numerical model. The PhyCNN paradigm also outperforms non-physics-guided neural networks.

SPIND: a reference-based auto-indexing algorithm for sparse serial crystallography data.

SPIND (sparse-pattern indexing) is an auto-indexing algorithm for sparse snapshot diffraction patterns ('stills') that requires the positions of only five Bragg peaks in a single pattern, when provided with unit-cell parameters. The capability of SPIND is demonstrated for the orientation determination of sparse diffraction patterns using simulated data from microcrystals of a small inorganic molecule containing three iodines, 5-amino-2,4,6-triiodoisophthalic acid monohydrate (I3C) [Beck & Sheldrick (2008 ▸), Acta Cryst. E64, o1286], which is challenging for commonly used indexing algorithms. SPIND, integrated with CrystFEL [White et al. (2012 ▸), J. Appl. Cryst. 45, 335-341], is then shown to improve the indexing rate and quality of merged serial femtosecond crystallography data from two membrane proteins, the human δ-opioid receptor in complex with a bi-functional peptide ligand DIPP-NH2 and the NTQ chloride-pumping rhodopsin (CIR). The study demonstrates the suitability of SPIND for indexing sparse inorganic crystal data with smaller unit cells, and for improving the quality of serial femtosecond protein crystallography data, significantly reducing the amount of sample and beam time required by making better use of limited data sets. SPIND is written in Python and is publicly available under the GNU General Public License from https://github.com/LiuLab-CSRC/SPIND.

ECOHYDRAULIC ANALYSIS OF FISH FATIGUE DATA

ABSTRACTSwimming speeds and endurance vary with species and body morphology, fish length, water temperature and other variables. Data on fish swimming performance are collected from various swim tests conducted in the laboratory or the field. Tests include stamina tunnels where fish are constrained, long pipes with pressure flow, open channels where fish are either constrained or free to volitionally swim and field studies with culverts or other hydraulic structures where fish passage is assessed. Studies on swimming performance and data availability on various species, mostly from stamina tunnels, have been increasing since the 1960s, with more emphasis on volitional channels since the 1990s. An extensive database on fish swimming performance was generated from the literature. Fish tested represent a broad range of species and individuals within many species. There are insufficient data for many species, and significant regressions for speed versus endurance are not available for these species. Fatigue curves of groups of species using an ecohydraulic approach with dimensionless quantities improved regressions when compared with body lengths per second versus time. Furthermore, it was found that the square root of body length is a better scale for fish speed. Although variability in swimming performance exists between species and individuals within a species, data analyses indicate broad similarities in relative performance for groups of species of similar morphology or swimming mode. The ecohydraulic approach allowed the use of limited data sets in analyses for groups of species. Significant speed–time regressions were developed for three groups, the Eel group, the Trout group and the Sturgeon group. Estimates of fish fatigue time with different confidence levels may be useful when considering physiological aspects in practical applications. Copyright © 2011 John Wiley & Sons, Ltd.

Image reconstructions in diffraction tomography from limited transmitted field data sets

Diffraction tomography reconstructions of objects from limited transmitted field data sets are discussed together with theoretical analyses and results of numerical experiments. It is shown that limited data sets, representing only a small part of the complete data sets, can be used for reconstructions in diffraction tomography with satisfactory accuracy. We also find that, in diffraction tomography based on the hybrid filtered backpropagation and the first-Rytov approximation, the use of limited data sets can provide a larger range of validity than the use of complete data sets, the reason being that limited data sets pose less-severe phase-unwrapping problems.

Effect of limited data sets in evaluating the scaling properties of spatially distributed data: an example from mining-induced seismic activity

The estimates of the scaling properties of various geophysical systems may be significantly affected by the use of limited data sets featuring finite numbers of data, finite sizes of study volumes, and measurement errors. These effects are illustrated through the spatial distribution of induced seismic activity in Creighton Mine (northern Ontario, Canada). The events studied occurred during a three-month period, March-May 1992, within a volume of approximate size 400 × 400 × 180 m3. Two data sets are considered, the first one consisting of the most accurately located micro-earthquakes (14338 events), and the second one including the portion of the first set that features the strongest microearthquakes (1654 events). The scaling properties of the spatial distribution of these events arc studied using generalized correlation integrals. From these, generalized correlation dimensions are estimated using the slope method. The dimension spectra are examined for the real data sets, randomly generated point sets featuring uniform and monofractal distributions and mimicking the limitations of the real data, and samples randomly extracted from the real data through a bootstrap procedure. The random simulations indicate that the uniform and monofractal random distributions can show spurious multifractality due only to the use of limited data sets. The re-sampling procedure demonstrates that is is possible to work effectively with small data sets. A comparison of the results from the real data, random point sets, and the re-sampled real data makes it possible to conclude that: (1) the bias in the estimates of the correlation dimensions from limited data sets car be readily evaluated, making it unnecessary to work with ever-increasing data sets; (2) correlation dimensions estimated from data sets featuring different limitations cannot be directly compared, neither is it recommended to assign specific physical meanings to their numerical values; (3) the strong multifractality suggested by the real dimension spectra in this study appears to be mainly spurious in character; (4) the spatial distribution of the larger microearthquakes, while different from a uniform distribution, could originate from a monofractal process; (5) the spatial distribution of the smaller microearthquakes is either monofractal or only weakly multifractal; and (6) small data sets can be effectively used to observe temporal variations in the scaling properties that may be associated with the occurrence of larger events.