Optimal Combination Of Features Research Articles

Enhancing Monthly Streamflow Prediction Using Meteorological Factors and Machine Learning Models in the Upper Colorado River Basin

Streamflow prediction is crucial for planning future developments and safety measures along river basins, especially in the face of changing climate patterns. In this study, we utilized monthly streamflow data from the United States Bureau of Reclamation and meteorological data (snow water equivalent, temperature, and precipitation) from the various weather monitoring stations of the Snow Telemetry Network within the Upper Colorado River Basin to forecast monthly streamflow at Lees Ferry, a specific location along the Colorado River in the basin. Four machine learning models—Random Forest Regression, Long short-term memory, Gated Recurrent Unit, and Seasonal AutoRegresive Integrated Moving Average—were trained using 30 years of monthly data (1991–2020), split into 80% for training (1991–2014) and 20% for testing (2015–2020). Initially, only historical streamflow data were used for predictions, followed by including meteorological factors to assess their impact on streamflow. Subsequently, sequence analysis was conducted to explore various input-output sequence window combinations. We then evaluated the influence of each factor on streamflow by testing all possible combinations to identify the optimal feature combination for prediction. Our results indicate that the Random Forest Regression model consistently outperformed others, especially after integrating all meteorological factors with historical streamflow data. The best performance was achieved with a 24-month look-back period to predict 12 months of streamflow, yielding a Root Mean Square Error of 2.25 and R-squared (R2) of 0.80. Finally, to assess model generalizability, we tested the best model at other locations—Greenwood Springs (Colorado River), Maybell (Yampa River), and Archuleta (San Juan) in the basin.

Using Machine Learning with Impulse Oscillometry Data to Develop a Predictive Model for Chronic Obstructive Pulmonary Disease and Asthma.

We aimed to develop and validate a machine learning model using impulse oscillometry system (IOS) profiles for accurately classifying patients into three assessment-based categories: no airflow obstruction, asthma, and chronic obstructive pulmonary disease (COPD). Our research questions were as follows: (1) Can machine learning methods accurately classify obstructive disease states based solely on multidimensional IOS data? (2) Which IOS parameters and modeling algorithms provide the best discrimination? We used data for 480 patients (240 with COPD and 240 with asthma) and 84 healthy individuals for training. Physiological and IOS parameters were combined into six feature combinations. The classification algorithms tested were logistic regression, random forest, neural network, k-nearest neighbor, and support vector machine. The optimal feature combination for identifying individuals without pulmonary obstruction, with asthma, or with COPD included 15 IOS and physiological features. The neural network classifier achieved the highest accuracy (0.786). For discriminating between healthy and unhealthy individuals, two combinations of twenty-three features performed best in the neural network algorithm (accuracy of 0.929). When distinguishing COPD from asthma, the best combination included 15 features and the neural network algorithm achieved an accuracy of 0.854. This study provides compelling technical evidence and clinical justifications for advancing IOS data-driven models to aid in COPD and asthma management.

Abstract ND09: Actinium-225 -PSMA-Trillium (BAY 3563254): Preclinical evaluation and clinical imaging study of a novel 225Ac-labeled PSMA-targeting small molecule triad for the treatment of mCRPC

Abstract With the approval of a PSMA-targeting small molecule labeled with the beta-emitter lutetium-177, PSMA is considered a clinically validated target for the treatment of mCRPC. Although effective, many patients treated with this therapeutic experience only transient clinical benefit, and G1/2 xerostomia is common due to salivary gland toxicity. Since the linear energy transfer and degree of induced DNA damage of alpha particles far exceeds that of beta particles, several investigational PSMA-targeting small molecules labeled with the alpha-emitter 225Ac have generated encouraging preliminary efficacy signals, but with severe xerostomia as a common dose-limiting toxicity. 225Ac-PSMA-Trillium is a novel PSMA-targeting radioligand consisting of a triad of important features: (1) a high-affinity PSMA inhibitor for specific tumor-targeting; (2) a customized albumin-binding domain designed to prolong plasma residence time and increase the therapeutic index by improving tumor uptake while reducing salivary gland uptake; and (3) a “macropa” chelator for efficient and stable radiolabeling with actinium-225. A series of compounds containing the same high-affinity PSMA-targeting moiety, but different chain lengths and albumin-binding domains, were evaluated to identify the optimal combination of features required to achieve the best therapeutic index. In vitro preclinical characterization was performed to test affinities for PSMA and albumin, compound stability and cell-killing potential in LNCaP and 22Rv1 CDX models as well as KuCaP-1 PDX model. Preclinical in vivo investigation of selected compounds showed increased plasma residence time, reduced uptake in normal organs, strong uptake and retention in PSMA-expressing tumors and robust anti-tumor activity. These analyses led to the selection of compounds for further evaluation in a Phase 0 imaging and PK study in mCRPC patients (NCT05773703). In this Phase 0 study, PSMA-Trillium compounds were labeled with indium-111 via a DOTA chelator and used as surrogates to facilitate longitudinal imaging. In patients, 111In-PSMA-Trillium demonstrated targeting to PSMA-expressing tumors as defined by PET/PSMA scans and a plasma residence time consistent with albumin-binding affinities. Importantly, 111In-PSMA-Trillium showed robust uptake and prolonged retention in tumors, and reduced uptake and relatively fast washout from salivary glands. Based on these preclinical data and clinical imaging results, a therapeutic first-in-human trial to evaluate 225Ac-PSMA-Trillium in patients with prostate cancer is anticipated to begin enrolling patients in the first half of 2024. Citation Format: Sabine Zitzmann-Kolbe, Jacob Hesterman, Patricia E. Cole, Nicolas Salem, Alex Papple, Michael Jeffers, Goekben Koca, Thorsten Poethko, Miroslav Dostalek, Shawn Hillier, John W. Babich. Actinium-225 -PSMA-Trillium (BAY 3563254): Preclinical evaluation and clinical imaging study of a novel 225Ac-labeled PSMA-targeting small molecule triad for the treatment of mCRPC [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(7_Suppl):Abstract nr ND09.

A Feature Selection Method Based on Relief Feature Ranking with Recursive Feature Elimination for the Inversion of Urban River Water Quality Parameters Using Multispectral Imagery from an Unmanned Aerial Vehicle

The timely monitoring of urban water bodies using unmanned aerial vehicle (UAV)-mounted remote sensing technology is crucial for urban water resource protection and management. Addressing the limitations of the use of satellite data in inferring the water quality parameters of small-scale water bodies due to their spatial resolution constraints and limited input features, this study focuses on the Zao River in Xi’an City. Leveraging UAV multispectral imagery, a feature selection method based on Relief Feature Ranking with Recursive Feature Elimination (Relief F-RFE) is proposed to determine the quality parameters of the typical urban pollution in water (dissolved oxygen (DO), total nitrogen (TN), turbidity, and chemical oxygen demand (COD). By constructing a potential feature set and utilizing optimal feature combinations, inversion models are developed for the four water quality parameters using three machine learning (ML) algorithms (Random Forest (RF), Support Vector Regression (SVR), Light Gradient Boosting Machine (LightGBM). The inversion accuracies of the different models are compared, and the spatial distribution of the four water quality parameters is analyzed. The results show that the models constructed based on UAV-based multispectral remote sensing imagery perform well in inferring the water quality parameters of the Zao River. The SVR algorithm, based on Relief F-RFE feature selection, achieves a higher accuracy, with RMSE values of 7.19 mg/L, 1.14 mg/L, 3.15 NTU, and 4.28 mg/L, respectively. The methods and conclusions of this study serve as a reference for research on the inversion of water quality parameters in urban rivers.

Conditional selection with CNN augmented transformer for multimodal affective analysis

AbstractAttention mechanism has been a successful method for multimodal affective analysis in recent years. Despite the advances, several significant challenges remain in fusing language and its nonverbal context information. One is to generate sparse attention coefficients associated with acoustic and visual modalities, which helps locate critical emotional semantics. The other is fusing complementary cross‐modal representation to construct optimal salient feature combinations of multiple modalities. A Conditional Transformer Fusion Network is proposed to handle these problems. Firstly, the authors equip the transformer module with CNN layers to enhance the detection of subtle signal patterns in nonverbal sequences. Secondly, sentiment words are utilised as context conditions to guide the computation of cross‐modal attention. As a result, the located nonverbal features are not only salient but also complementary to sentiment words directly. Experimental results show that the authors’ method achieves state‐of‐the‐art performance on several multimodal affective analysis datasets.

Machine learning-driven optimization and application of bimetallic catalysts in peroxymonosulfate activation for degradation of fluoroquinolone antibiotics

In this research, the XGBoost (XGB) and CatBoost (CB) algorithms were utilized to examine the experimental refinement and deployment of bimetallic catalysts for the activation of peroxymonosulfate in the removal of fluoroquinolone antibiotics from aqueous solutions. For the XGB model, shapley additive explanations analysis was utilized to provide model interpretability, identifying the features with significant impact on model prediction. By integrating multiple optimization techniques, such as particle swarm optimization and differential evolution, we successfully determined the optimal feature combination for the XGB model, providing accurate guidance for degradation experiments. Under the designed experimental conditions, the two sets of materials could degrade 93% of ciprofloxacin and 92% of norfloxacin within 30 and 27 min, respectively, with the discrepancies from model-predicted outcomes all within 3%. Concurrently, through electron paramagnetic resonance testing and quenching experiments, we explored the role of reactive oxygen species in the catalytic degradation process, with the experimental evidence demonstrating a 100% prediction accuracy for the CB model. The experimental results underscored the substantial potential of machine learning in the optimization design of environmental engineering.

Estimating Winter Wheat Plant Nitrogen Content by Combining Spectral and Texture Features Based on a Low-Cost UAV RGB System throughout the Growing Season

As prior information for precise nitrogen fertilization management, plant nitrogen content (PNC), which is obtained timely and accurately through a low-cost method, is of great significance for national grain security and sustainable social development. In this study, the potential of the low-cost unmanned aerial vehicle (UAV) RGB system was investigated for the rapid and accurate estimation of winter wheat PNC across the growing season. Specifically, texture features were utilized as complements to the commonly used spectral information. Five machine learning regression algorithms, including support vector machines (SVMs), classification and regression trees, artificial neural networks, K-nearest neighbors, and random forests, were employed to establish the bridge between UAV RGB image-derived features and ground-truth PNC, with multivariate linear regression serving as the reference. The results show that both spectral and texture features had significant correlations with ground-truth PNC, indicating the potential of low-cost UAV RGB images to estimate winter wheat PNC. The H channel, S4O6, and R_SE and R_EN had the highest correlation among the spectral indices, Gabor texture features, and grey level co-occurrence matrix texture features, with absolute Pearson’s correlation coefficient values of 0.63, 0.54, and 0.69, respectively. When the texture features were used together with spectral indices, the PNC estimation accuracy was enhanced, with the root mean square error (RMSE) decreasing from 2.56 to 2.24 g/kg, for instance, when using the SVM regression algorithm. The SVM regression algorithm with validation achieved the highest estimation accuracy, with a coefficient of determination (R2) of 0.62 and an RMSE of 2.15 g/kg based on the optimal feature combination of B_CON, B_M, G_DIS, H, NGBDI, R_EN, R_M, R_SE, S3O7, and VEG. Overall, this study demonstrated that the low-cost UAV RGB system could be successfully used to map the PNC of winter wheat across the growing season.

Fine Classification of Urban Tree Species Based on UAV-Based RGB Imagery and LiDAR Data

Rapid and accurate classification of urban tree species is crucial for the protection and management of urban ecology. However, tree species classification remains a great challenge because of the high spatial heterogeneity and biodiversity. Addressing this challenge, in this study, unmanned aerial vehicle (UAV)-based high-resolution RGB imagery and LiDAR data were utilized to extract seven types of features, including RGB spectral features, texture features, vegetation indexes, HSV spectral features, HSV texture features, height feature, and intensity feature. Seven experiments involving different feature combinations were conducted to classify 10 dominant tree species in urban areas with a Random Forest classifier. Additionally, Plurality Filling was applied to further enhance the accuracy of the results as a post-processing method. The aim was to explore the potential of UAV-based RGB imagery and LiDAR data for tree species classification in urban areas, as well as evaluate the effectiveness of the post-processing method. The results indicated that, compared to using RGB imagery alone, the integrated LiDAR and RGB data could improve the overall accuracy and the Kappa coefficient by 18.49% and 0.22, respectively. Notably, among the features based on RGB, the HSV and its texture features contribute most to the improvement of accuracy. The overall accuracy and Kappa coefficient of the optimal feature combination could achieve 73.74% and 0.70 with the Random Forest classifier, respectively. Additionally, the Plurality Filling method could increase the overall accuracy by 11.76%, which could reach 85.5%. The results of this study confirm the effectiveness of RGB imagery and LiDAR data for urban tree species classification. Consequently, these results could provide a valuable reference for the precise classification of tree species using UAV remote sensing data in urban areas.

Multi-channel feature extraction for virtual histological staining of photon absorption remote sensing images

Accurate and fast histological staining is crucial in histopathology, impacting diagnostic precision and reliability. Traditional staining methods are time-consuming and subjective, causing delays in diagnosis. Digital pathology plays a vital role in advancing and optimizing histology processes to improve efficiency and reduce turnaround times. This study introduces a novel deep learning-based framework for virtual histological staining using photon absorption remote sensing (PARS) images. By extracting features from PARS time-resolved signals using a variant of the K-means method, valuable multi-modal information is captured. The proposed multi-channel cycleGAN model expands on the traditional cycleGAN framework, allowing the inclusion of additional features. Experimental results reveal that specific combinations of features outperform the conventional channels by improving the labeling of tissue structures prior to model training. Applied to human skin and mouse brain tissue, the results underscore the significance of choosing the optimal combination of features, as it reveals a substantial visual and quantitative concurrence between the virtually stained and the gold standard chemically stained hematoxylin and eosin images, surpassing the performance of other feature combinations. Accurate virtual staining is valuable for reliable diagnostic information, aiding pathologists in disease classification, grading, and treatment planning. This study aims to advance label-free histological imaging and opens doors for intraoperative microscopy applications.

A multi-feature stock price prediction model based on multi-feature calculation, LASSO feature selection, and Ca-LSTM network

This paper addresses the crucial realm of stock price prediction, highly coveted by individual investors and institutions for its substantial economic implications. The inherent non-stationary and intricate nature of stock market fluctuations, coupled with real-time transactions, poses a formidable challenge for accurate and swift prediction. Unlike prevailing research that predominantly focuses on forecasting methods, our novel approach places a paramount emphasis on processing original data, introducing 57 technical indicators to better represent economic aspects for stock price prediction. Signifying the importance of each feature, we employ the LASSO algorithm to derive an optimal feature combination. Additionally, our methodology utilizes the Ca-LSTM (cascade long short-term memory) technique, enhancing information extraction from individual features. Experimental results, gauged by mean error, underscore the superiority of the Ca-LSTM model over other time series prediction models and conventional long short-term memory approaches. Notably, our model's integration with the accumulation-based VMD-LSTM model demonstrates enhanced forecasting accuracy. This proposed method holds considerable potential to refine stock price prediction, thereby delivering heightened value to investors in the dynamic financial landscape.

High-precision retrieval of offshore sea surface temperature: A machine learning framework based on MODIS and in-situ measurements

High-precision sea surface temperature (SST) data is essential for monitoring and management of offshore environment. SST retrievals from thermal infrared remote sensing have the advantages in time frequency and space coverage over measurements from ships and buoys. However, traditional SST retrieval models are not fully applicable for offshore waters where seawater conditions change a lot, due to simple functional forms and empirical fitting of model parameters. Machine learning (ML) is introduced because it doesn’t need to define specific model rules or unknown parameters, which is more intelligent to deal with complex retrieval problems in a data-driven way. But the applicability of ML has not been investigated in the North Yellow Sea of China, and the selection of driving features for model construction has not been fully discussed. Here, we designed a high-precision SST retrieval framework based on random forest (RF) with optimal feature combination using MODIS data and in-situ SST measurements during 2013–2020 in the northern part of the North Yellow Sea. The performances of different feature combinations were assessed based on model testing accuracies and the RF output feature importance. Compared with previous ML-based SST retrieval studies, we added two time-related features (i.e., day of the year and month) and one feature as initial temperature (i.e., SST retrieval from split-window algorithm), which reduced the testing error by around 15 % and 11 %, respectively. The retrieval results were validated with in-situ measurements, showing that the RF model achieved better accuracies (R2 = 0.987, SD = 0.842 °C, MAE = 0.675 °C, RMSE = 0.841 °C) than the improved split-window algorithm (named as ISW) and deep neural network. Compared with ISW, the spatial distribution of RF retrieval map exhibited similar variation across four seasons, mainly differed in the coastal areas or at the formation of temperature fronts. Note that the constructed RF model was still reliable when applied to the independent samples in 2021. This study contributes to improve the accuracy of offshore SST retrieval under different seawater conditions and provides references for SST monitoring.

A novel profit-based validity index approach for feature selection in credit risk prediction

<abstract> <p>Establishing a reasonable and effective feature system is the basis of credit risk early warning. Whether the system design is appropriate directly determines the accuracy of the credit risk evaluation results. In this paper, we proposed a feature system through a validity index with maximum discrimination and commercial banks' loan profit maximization. First, the first objective function is the minimum validity index constructed by the intra-class, between-class, and partition coefficients. The maximum difference between the right income and wrong cost is taken as the second objective function to obtain the optimal feature combination. Second, the feature weights are obtained by calculating the change in profit after deleting each feature with replacement to the sum of all change values. An empirical analysis of 3, 425 listed companies from <italic>t</italic>-1 to <italic>t</italic>-5 time windows reveals that five groups of feature systems selected from 614 features can distinguish between defaults and non-defaults. Compared with 14 other models, it is found that the feature systems can provide at least five years' prediction and enable financial institutions to obtain the maximum profit.</p> </abstract>

Cross-subject EEG-based emotion recognition through dynamic optimization of random forest with sparrow search algorithm.

The objective of EEG-based emotion recognition is to classify emotions by decoding signals, with potential applications in the fields of artificial intelligence and bioinformatics. Cross-subject emotion recognition is more difficult than intra-subject emotion recognition. The poor adaptability of classification model parameters is a significant factor of low accuracy in cross-subject emotion recognition. We propose a model of a dynamically optimized Random Forest based on the Sparrow Search Algorithm (SSA-RF). The decision trees number (DTN) and the leave minimum number (LMN) of the RF are dynamically optimized by the SSA. 12 features are used to construct feature combinations for selecting the optimal feature combination. DEAP and SEED datasets are employed for testing the performance of SSA-RF. The experimental results show that the accuracy of binary classification is 76.81% on DEAP, and the accuracy of triple classification is 75.96% on SEED based on SSA-RF, which are both higher than that of traditional RF. This study provides new insights for the development of cross-subject emotion recognition, and has significant theoretical value.

Determining the best feature combination through text and probabilistic feature analysis for GPT-2-based mobile app review detection

Mobile apps, used by many people worldwide, have become an essential part of life. Before using a mobile app, users judge the reliability of apps according to their reviews. Therefore, app reviews are essential components of management for companies. Unfortunately, some fake reviewers write negative reviews for competing apps. Moreover, artificial intelligence (AI)-based macro bot programs that generate app reviews have emerged and can create large numbers of reviews with malicious purposes in a short time. One notable AI technology that can generate such reviews is Generative Pre-trained Transformer-2 (GPT-2). The reviews generated by GPT-2 use human-like grammar; therefore, it is difficult to detect them with only text mining techniques, which use tools like part-of-speech (POS) tagging and sentiment scores. Thus, probability-based sampling techniques in GPT-2 must be used. In this study, we identified features to detect reviews generated by GPT-2 and determined the optimal feature combination for improving detection performance. To achieve this, based on the analysis results, we built a training dataset to find the best feature combination for detecting the generated reviews. Various machine learning models were then trained and evaluated using this dataset. As a result, the model that used both text mining and probability-based sampling techniques detected generated reviews more effectively than the model that used only text mining techniques. This model achieved a top classification accuracy of 90% and a macro F1 of 0.90. We expect the results of this study to help app developers maintain a more stable mobile app ecosystem.Graphical abstract

URBAN GREEN SPACE IDENTIFICATION BY FUSING SATELLITE IMAGES FROM GF-2 AND SENTINEL-2

Abstract. This paper combines class hierarchy construction and feature-preferred random forest classification method, and the classification results are the best. In the urban center area with complex features, this method can be used to extract small and complex features more accurately, and for urban green spaces, small auxiliary green spaces between houses can be accurately extracted. This method first constructs a class hierarchy of four sizes, and then extracts different features from simple to complex, from large to small, and classifies them by membership function for the direct selection feature rules of easily extracted features. For the subdivision of green space and the classification of features in central complex areas, feature optimization is carried out, and the optimal feature combination is selected and then extreme random tree (ERT) classification is performed. The classification accuracy is the best 89.5%, and the classification results are analyzed correctly, which correctly distinguishes the smaller land categories in the central area, reduces the misclassification of grassland and agricultural land, and the classification results are optimal.

Aboveground Forest Biomass Estimation Using Tent Mapping Atom Search Optimized Backpropagation Neural Network with Landsat 8 and Sentinel-1A Data

Accurate forest biomass estimation serves as the foundation of forest management and holds critical significance for a comprehensive understanding of forest carbon storage and balance. This study aimed to integrate Landsat 8 OLI and Sentinel-1A SAR satellite image data and selected a portion of the Shanxia Experimental Forest in Jiangxi Province as the study area to establish a biomass estimation model by screening influencing factors. Firstly, we extracted spectral information, vegetation indices, principal component features, and texture features within 3 × 3-pixel neighborhoods from Landsat 8 OLI. Moreover, we incorporated Sentinel-1’s VV (vertical transmit–vertical receive) and VH (vertical transmit–horizontal receive) polarizations. We proposed an ensemble AGB (aboveground biomass) model based on a neural network. In addition to the neural network model, namely the tent mapping atom search optimized BP neural network (Tent_ASO_BP) model, partial least squares regression (PLSR), support vector machine (SVR), and random forest (RF) regression prediction techniques were also employed to establish the relationship between multisource remote sensing data and forest biomass. Optical variables (Landsat 8 OLI), SAR variables (Sentinel-1A), and their combinations were input into the four prediction models. The results indicate that Tent_ ASO_ BP model can better estimate forest biomass. Compared to pure optical or single microwave data, the Tent_ASO_BP model with the optimal combination of optical and microwave input features achieved the highest accuracy. Its R2 was 0.74, root mean square error (RMSE) was 11.54 Mg/ha, and mean absolute error (MAE) was 9.06 Mg/ha. Following this, the RF model (R2 = 0.54, RMSE = 21.33 Mg/ha, MAE = 17.35 Mg/ha), SVR (R2 = 0.52, RMSE = 17.66 Mg/ha, MAE = 15.11 Mg/ha), and PLSR (R2 = 0.50, RMSE = 16.52 Mg/ha, MAE = 12.15 Mg/ha) models were employed. In conclusion, the BP neural network model improved by tent mapping atom search optimization algorithm significantly enhanced the accuracy of AGB estimation in biomass studies. This will provide a new avenue for large-scale forest resource surveys.

Predicting Potential Risk: Cerebral Stroke via Regret Minimization

Objective. The data processing of medical test report has always been one of the important contents in biological information domain, especially the process of extracting the effective information from the report so as to assist doctors with the correct medical plan. Usual methods neglect the implicit relationship between features. More features are generally not a better choice because more noise is generated between feature combinations. We propose a practical feature selection strategy RMFS, which aims to select the optimal combination of features. Materials and Methods. Based on the above situation, in this paper, 64 features are extracted from a real medical test report dataset for stroke and feature selection is defined as a reinforcement learning problem to optimize the feature combination by minimizing regret. We select three current mainstream feature selection methods and conduct comparative experiments. Results. We processed and completed a dataset derived from real medical test reports of stroke. We redefine the feature selection problem as a reinforcement learning problem and propose an optimization strategy based on regret minimization and train weight parameters in a DQN network. Experimental results demonstrate that our method can identify feature combinations with higher prediction accuracy. Discussion. RMFS shows a strong robustness to the randomness of the environment and has high computational efficiency and accuracy. Compared with the previous feature selection methods, our method yields superior results. Conclusion. The experimental results demonstrate that our method can obtain a more accurate prediction rate under the same feature scale and we can achieve baseline performance with fewer features.

Unsupervised feature selection by learning exponential weights

Unsupervised feature selection has gained considerable attention for extracting valuable features from unlabeled datasets. Existing approaches typically rely on sparse mapping matrices to preserve local neighborhood structures. However, this strategy favors large-weight features, potentially overlooking smaller yet valuable ones and distorting data distribution and feature structure. Besides, some methods focus on local structure information, failing to explore global information. To address these limitations, we introduce an exponential weighting mechanism to induce a rational feature distribution and explore data structure in the feature subspace. Specifically, we propose a unified framework incorporating local structure learning and exponentially weighted sparse regression for optimal feature combinations, preserving global and local information. Experimental results demonstrate the superiority of our approach over existing unsupervised feature selection methods.

Comparison of 2D and 3D vegetation species mapping in three natural scenarios using UAV-LiDAR point clouds and improved deep learning methods

Collaboration between Light Detection and Ranging (LiDAR) point clouds and deep learning has been proven to be an effective approach for vegetation mapping. Current studies have predominantly focused on 2D vegetation mapping, whereas 3D mapping, which directly classifies point clouds at point level, offers a more comprehensive understanding of the stratified structural information of vegetation. However, there is a lack of research on 3D vegetation species mapping, and the disparities between 2D and 3D mapping in natural scenarios remain unclear. To resolve these issues, we compared the deep learning performance of 2D and 3D vegetation species mapping across three distinct natural scenes: karst wetland, mangrove forest, and hill forest. In addition, the 2D and 3D mapping in natural scenes are adversely affected by the elevated channel count of LiDAR-derived features and the extreme category imbalance in point cloud. To mitigate these challenges, we propose a novel Multi-resolution Feature Selection Network (MrFSNet) to select optimal feature combinations at different scales for better 2D mapping performance. Additionally, we introduce a novel Dynamic Weighted Sampling (DWS) strategy, which is combined with KPConv to address the extreme category imbalance present in 3D mapping. Results indicate that: (1) 3D vegetation species mapping exhibited the highest performance, achieving an mF1 of 89.78% for karst wetland, 92.25% for mangrove forest, and 92.05% for hill forest. (2) 3D mapping outperformed 2D mapping, improving mF1 by 3.43% to 27.08%. (3) MrFSNet adaptively extracted optimal features at various scales and performed well with the limited training data in 2D vegetation mapping, resulting in a 1.66%–18.46% higher mF1 than that of Swin Transformer. (4) DWS effectively resolved the extreme category imbalance problem and produced 1.28%–2.80% higher mF1 than the non-DWS version in 3D vegetation mapping.

Discriminating the default risk of small enterprises: Stacking model with different optimal feature combinations

Small enterprise default discrimination establishes a risk discriminant model based on financial data, non-financial data, and external macro conditions of small enterprises to obtain their default discriminant. This study constructs the final default discriminant model, taking the predicted default probability vectors from the first modelling as independent variables. From the 2e feature combinations composed of e features, it maximises the default discriminant precision of the same training sample. Accordingly, the study inversely infers three optimal feature combinations corresponding to three models, including logistic regression, support vector machine, and linear discriminant analysis, ensuring the discrimination accuracy of the first modelling by the stacking method. Moreover, five features—industry prosperity index, EBITDA margin, current assets turnover ratio, net profit, and per capita disposable income of urban residents—account for 11% of the features in the optimal feature combination, but their importance accounts for 63%; thus, they are critical to the default risk of small enterprises. Further, the macro features significantly influence the default risk of small enterprises. For example, the importance of four macro features—industry prosperity index, consumer price index, per capita disposable income of urban residents, and Engel coefficient—accounts for 26.26% of the features in the optimal feature combination. Notably, the importance of the ‘industry prosperity index’, which is the greatest influencing factor in the feature combination, accounts for 17.63%. Ultimately, the discrimination accuracy of the proposed model is better than that of the seven classical default discriminant models.