Soft reordering one-dimensional convolutional neural network for credit scoring

Qian, Hongyi; Ma, Ping; Gao, Songfeng; Song, You

doi:10.1016/j.knosys.2023.110414

Cited by 15 publications

(5 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…One-dimensional convolutional neural networks are commonly used to process data that is temporal in nature, such as audio, text, and time series [27]. In a one-dimensional convolutional neural network, the input data is a one-dimensional vector and the convolutional kernel moves only on this vector, hence the name one-dimensional convolution [28].…”

Section: One-dimensional Convolutional and Receptive Fieldmentioning

confidence: 99%

A receptive field transfer strategy via layer-aligned distillation learning for fault signal denoising

Pu,

Zhang,

2024

Meas. Sci. Technol.

View full text Add to dashboard Cite

To improve fault diagnosis performance in complex noise environments, effective signal denoising techniques are necessary. However, traditional denoising methods have proven inadequate for multivariate fault signal denoising, neglecting the correlation among these signals. To this end, we propose a novel denoising module, inspired by traditional signal decomposition and reconstruction methods. Furthermore, to enhance the performance of proposed denoising module, we consider the influence of the receptive field and develop a receptive field transfer strategy using layer-aligned distillation learning. The experiments demonstrate that our approach effectively balances the denoising performance and computational load, offering a novel strategy for developing high-performance denoising networks. What's more, our strategy reduces the difficulty for fault diagnosis tasks under complex noise environments.

show abstract

Section: One-dimensional Convolutional and Receptive Fieldmentioning

confidence: 99%

A receptive field transfer strategy via layer-aligned distillation learning for fault signal denoising

Pu,

Zhang,

2024

Meas. Sci. Technol.

View full text Add to dashboard Cite

show abstract

“…To enable the convolutional kernel to more effectively extract tabular features, referring to Qians work [39] on credit scoring, a direct approach that reshapes the input data into a multichannel image format is adopted. Subsequently, we employ a fully connected (FC) layer to learn the correct ordering through backpropagation, thereby addressing this challenge.…”

Section: A Soft Sortingmentioning

confidence: 99%

Enhancing Concrete Creep Prediction With Deep Learning: A Soft-Sorted One-Dimensional CNN Approach

Li,

Zhang,

Zhang

2023

IEEE Access

View full text Add to dashboard Cite

Creep is a long-term property of concrete that poses a significant challenge to structural engineering. The research content of this paper concerns how to accurately predict the creep behavior of concrete. The overall purpose is to develop a robust predictive model based on the Northwestern University (NU) database that overcomes the challenges encountered when handling tabular creep data, including the lack of spatial correlations between features and the presence of noise. To achieve this goal, a novel deep learning methodology that leverages convolutional neural networks (CNNs) is introduced. First, the data preprocessing phase involves feature selection and K-nearest neighbor-based data filling to enhance the quality of the given data. A soft-sorted 1D CNN is introduced to transform tabular data into multichannel images, thereby taking advantage of spatial locality and feature sorting. Additionally, this model incorporates noise reduction mechanisms through residual network encoders, achieving enhanced data clarity and model accuracy. The mean absolute error (MAE) and root mean square deviation (RMSE) of the model increase from 7.48 and 10.76 to 5.83 and 8.88, respectively. The major findings of this research indicate a significant predictive accuracy improvement over the existing methods. The model effectively captures the complex relationships between concrete creep and a range of influencing factors, enabling precise and quantitative predictions to be obtained. These findings offer deep insights for the field of civil engineering, enabling enhanced structural design, safety, and performance prediction in concrete applications. This paper contributes to the ongoing exploration of advanced deep learning techniques for addressing complex challenges in the domains of concrete mechanics and structural engineering.

show abstract

“…Techniques that account for deep learning still rely on subjective methods, such as saliency maps, which have a limited explanatory capability for unstructured data, such as images. On the other hand, significant progress has been made in making structured data explainable with deep learning [24].…”

Section: Introductionmentioning

confidence: 99%

Why Do Tree Ensemble Approximators Not Outperform the Recursive-Rule eXtraction Algorithm?

Onishi,

Nishimura,

Fujimura

et al. 2024

MAKE

View full text Add to dashboard Cite

Although machine learning models are widely used in critical domains, their complexity and poor interpretability remain problematic. Decision trees (DTs) and rule-based models are known for their interpretability, and numerous studies have investigated techniques for approximating tree ensembles using DTs or rule sets, even though these approximators often overlook interpretability. These methods generate three types of rule sets: DT based, unordered, and decision list based. However, very few metrics exist that can distinguish and compare these rule sets. Therefore, the present study proposes an interpretability metric to allow for comparisons of interpretability between different rule sets and investigates the interpretability of the rules generated by the tree ensemble approximators. We compare these rule sets with the Recursive-Rule eXtraction algorithm (Re-RX) with J48graft to offer insights into the interpretability gap. The results indicate that Re-RX with J48graft can handle categorical and numerical attributes separately, has simple rules, and achieves a high interpretability, even when the number of rules is large. RuleCOSI+, a state-of-the-art method, showed significantly lower results regarding interpretability, but had the smallest number of rules.

show abstract

Soft reordering one-dimensional convolutional neural network for credit scoring

Cited by 15 publications

References 33 publications

A receptive field transfer strategy via layer-aligned distillation learning for fault signal denoising

A receptive field transfer strategy via layer-aligned distillation learning for fault signal denoising

Enhancing Concrete Creep Prediction With Deep Learning: A Soft-Sorted One-Dimensional CNN Approach

Why Do Tree Ensemble Approximators Not Outperform the Recursive-Rule eXtraction Algorithm?

Contact Info

Product

Resources

About