Search Results (271)

Industrial energy consumption in shift-based manufacturing exhibits pronounced temporal asymmetry—here defined as direction-dependent conditional dynamics in which the transition from production to shutdown states follows a systematically different temporal trajectory than the reverse transition. At the facility studied, this asymmetry also manifests in the marginal distribution of hourly consumption values: pooling all 4724 observations yields a bimodal, right-skewed histogram (skewness ≈ −0.4) comprising two sub-populations corresponding to production hours (14–19 kWh/h) and shutdown hours (0–2 kWh/h). Although individual hourly observations are serially dependent and therefore not i.i.d., the marginal distributional shape is consequential because ARIMA-class models assume approximately Gaussian innovations, and residuals from models fit to this bimodal series inherit its non-Gaussianity. More fundamentally, the conditional distribution P(E_t|E_{t − 1}, …) is direction-dependent: the production-to-shutdown transition is abrupt (1–2 h, 18:00–20:00), while the shutdown-to-production ramp is slower and more variable (2–4 h, 05:00–07:00). Symmetric ARMA models, applying identical autoregressive coefficients regardless of transition direction, cannot represent this directional asymmetry, rendering their assumptions and associated error metrics structurally unreliable for this class of data. This paper addresses this asymmetry directly by presenting and evaluating two hybrid forecasting architectures—Prophet+LSTM and SARIMA+LSTM—for 24 h-ahead energy prediction at an industrial bread factory in Kazakhstan, instrumented with 15 IoT energy meters. The two-stage design exploits the complementary asymmetry-handling properties of each component: the statistical model (Prophet or SARIMA) captures deterministic seasonal structure, while the LSTM corrects asymmetric residuals that the statistical model systematically misrepresents. In a rigorous 14-day holdout evaluation, Prophet+LSTM achieves an MAE of 3.39 kWh—outperforming the Seasonal Naïve baseline by 12.3% and reducing Prophet-alone error by 32.7%—with statistical significance at the 10% level confirmed via Diebold–Mariano testing (DM = +1.747, p = 0.081). The LSTM residual correction reduces Prophet’s systematic negative bias by 69% (from −3.60 to −1.13 kWh), as confirmed by ablation testing. In eight weeks of production operation with incremental retraining, MAE improved 35% (7.02 → 4.58 kWh). These results demonstrate that explicitly modelling temporal asymmetry through hybrid statistical-neural architectures substantially improves industrial energy forecasting accuracy. Full article

Feature selection represents a critical step in developing accurate and interpretable models for early breast cancer detection. Despite extensive research in the field of mammographic image analysis, no consensus has yet been reached on the optimal feature subsets that distinguish normal from malignant tissues. To address this gap, the present study aims to identify the most discriminative and significant features through a comprehensive multi-criterion selection framework. The aim is to integrate, as new frameworks, different combinations of t-test, ANOVA, Mutual Information (MI), and Equal Grouping Methods (EGM) to rank 19 linear and nonlinear features extracted from mammographic images. The objective is to maximize feature relevance while minimizing redundancy and enhancing diagnostic and healthcare systems. Linear features were assessed alongside nonlinear descriptors. A framework combining t-test, ANOVA, and EGM, guided by MI relevance, was employed to balance feature contributions across categories. The experimental results demonstrated that hybrid feature selection significantly enhanced diagnostic accuracy using optimal linear and nonlinear attributes. The optimization results suggested using a hybrid of six linear and eight nonlinear features. Linear features were highly accurate for detecting cancer. Haralick entropy obtained the highest average accuracy and performance, 94.14% and 93.45%; followed by kurtosis, 93.49% and 92.59%; perimeter irregularity, 93.43% and 92.65%; skewness, 93.01% and 92.25%; and volume/area, 92.82% and 91.92%. Despite the reliable discriminative power of linear descriptors, their overall effectiveness in representing intricate tissue characteristics was limited. The comparison of statistical characteristics shows a distinct performance benefit of nonlinear descriptors over linear ones for detecting breast cancer. Nonlinear descriptors, however, showcased higher accuracy and performance, with an average accuracy of 97.81% in contrast to 94.43% for linear approaches. Local phase congruency achieved the top average accuracy and performance, 97.81% and 96.61%, respectively; succeeded by wavelet entropy, 97.62% and 96.42%; Laplacian spectrum features, 97.52% and 96.32%; nonlinear diffusion, 97.10% and 95.90%; and clustering coefficient, 96.70% and 95.50%; then Shannon, Tsallis, and Rényi entropies. The results indicate that statistically validated nonlinear characteristics significantly outperform linear ones across accuracy and performance measures. Their combination significantly improves the strength and discriminative power of computer-assisted breast cancer diagnostic systems, affirming their suitability for integration into sophisticated machine learning and deep learning models. The results also show that the new multi-criterion framework’s early detection performance surpassed that of the statistical and deep learning models explored, with an average of 98.6% accuracy, 98% sensitivity, 98.9% precision, and 98.4% F1 score of early detection of breast cancer. The incorporation of statistically validated nonlinear descriptors, particularly local phase congruency and wavelet entropy, improves the discriminative ability, robustness, and clinical understanding of breast cancer computer-assisted diagnostic systems. Overall, the proposed framework confirms that integrating hybrid features substantially enhances robustness and plays a pivotal role in computer-assisted breast cancer detection. These selected features may be fed to more advanced algorithms in the future, potentially yielding improved performance. Full article

This study investigates data augmentation as a strategy for addressing dataset scarcity in Internet of Medical Things (IoMT) cybersecurity and improving intrusion-detection system performance. Four augmentation methods—Rule-Based, Tabular Variational Autoencoder (TVAE), Conditional Tabular Generative Adversarial Network (CTGAN), and Gaussian Copula—were applied to two publicly available IoMT datasets (ECU-IoHT and WUSTL-EHMS) to generate augmented training data with differing class distributions and feature characteristics. Eleven machine learning algorithms were evaluated using Matthews Correlation Coefficient (MCC), F1-score, accuracy, and error-based metrics. Results showed consistent performance improvements across all evaluated models relative to the baseline datasets. The Rule-Based method produced the strongest overall results, achieving the highest MCC (0.9757), F1-score (99.19%), and accuracy (99.18%) with LightGBM, alongside low false-positive and false-negative rates. Among the generative approaches, TVAE delivered the strongest overall practical performance (F1-score = 96.94%, accuracy = 96.92%), while CTGAN achieved a marginally higher MCC (0.9047) and also produced competitive results with balanced class representation. Gaussian Copula generated the weakest overall outcomes, primarily due to highly skewed class distributions. Traditional models, such as Logistic Regression and Naive Bayes, recorded the largest relative gains, indicating that augmentation can substantially improve simpler classifiers in data-scarce environments. Overall, the findings demonstrate that augmentation quality depends not only on dataset expansion, but also on preserving class balance, feature diversity, and realistic traffic relationships. These results provide practical guidance for strengthening IoMT intrusion-detection systems in healthcare environments. Full article

This study proposes and validates a framework that integrates Grey Wolf Optimization (GWO) with Extreme Gradient Boosting (XGBoost) for estimating the Poisson’s ratio of auxetic structures. First, for 320 models derived from Computer-Aided Design-based (CAD-based) unit-cell designs, a systematic sweep of diameter and cellular dimensions was conducted to obtain porosity coverage in the 45–85% range. Subsequently, elastic modulus and Poisson’s ratio were computed via finite element analysis (FEA) at three mesh resolutions (0.20/0.25/0.30 mm), and relationships between design variables and outputs were examined using correlation heatmaps and Locally Weighted Scatterplot Smoothing (LOWESS) curves. GWO optimized the XGBoost hyperparameters through a multi-band narrowed search strategy; performance was evaluated using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Squared Error (MSE), and Coefficient of Determination ($R^2$) metrics, as well as residual diagnostics and Ground Truth–Prediction alignments for Poisson’s ratio. Across all configurations, $R^2\ge 0.994$ and absolute errors are on the order of ∼${10}^{-3}$; the 0.25 mm mesh stands out in terms of overall balance with the lowest squared-error profile and the highest $R^2$, the 0.30 mm mesh is practically equivalent in terms of MAE, and the 0.20 mm mesh is comparatively weaker. Residual diagnostics—comprising a pattern-free cloud around zero, slight right-skewness, and limited heteroskedasticity—indicate low bias and no substantive model-specification issues. The findings align with physical insight, confirming that Poisson’s ratio shifts toward more negative values as porosity increases and toward less negative values as diameter increases. The proposed GWO–XGBoost framework provides a reliable pre-screening tool for rapid design exploration and Poisson’s-ratio-targeted optimization, with the potential to reduce the need for additional FEA simulations and experimental iterations during early-stage design. Full article

This study proposes a transparent, data-driven framework for behavior recognition based exclusively on IMU measurements, hypothesizing that vehicular jerk-based features can help in differentiating driving behavior. Unlike studies relying on direct jerk values, our approach derives novel findings from jerk-based features. For rolling windows of 300 samples, a comprehensive set of statistical and dynamic descriptors is extracted, including amplitude, variance, standard deviation, coefficient of variation, standard error, skewness, and kurtosis, as well as jerk-based features such as jerk_std, jerk_variance, jerk_amplitude, and jerk_spikes. Statistical analysis is used to identify features with strong discriminative power. The selected features are used to compute the Driving Score (DS) and, along with the Kernel Density Estimation (KDE) and associated statistics, provide a driver’s profile. Low DS values are consistently associated with increased jerk variability, whereas high DS values correspond to smoother and more controlled motion profiles. The robustness of the proposed framework is evaluated using several machine learning classifiers as baselines, with the jerk-based features as inputs. For the aggressive driver class, the Driving Behavior Score (DBS) model reports a Recall of 0.952 and an F1 of 0.925. For the normal driver class, the DBS model reports a Recall of 0.839 and an F1 of 0.879. The model has a total accuracy of 0.907. Also, Logistic Regression and ensemble models like Extreme Gradient Boosting (XGB) and Random Forest (RF) perform well. The proposed framework offers an explainable, computationally efficient alternative to conventional machine-learning classifiers for identifying aggressive drivers. It relies on lightweight statistical computations being suitable for real-time implementation. Full article

We develop a semi-parametric framework for representing discrete probability mass functions through orthogonal polynomial representations. Classical count models, such as the Poisson and negative binomial distributions, impose restrictive structural assumptions that often fail to accommodate empirical features including heavy overdispersion, multimodality, and nonstandard tail behavior. To address these limitations, we introduce a linear-tilt model constructed from orthonormal polynomial systems associated with Poisson and negative binomial baselines, namely the Charlier and Meixner families. The proposed representation improves the baseline distribution using additional information from empirical moments. This allows the distribution to flexibly adjust its shape, capturing differences in skewness and kurtosis. We establish theoretical properties of the expansion within a weighted Hilbert space formulation, where the coefficients arise as orthogonal projections that can be expressed as expectations of the corresponding polynomial basis functions. In addition, we analyze approximation behavior and provide numerical bounds on the resulting numerical error and convergence properties of truncated approximations. The practical relevance of the proposed methodology is illustrated through applications to several empirical datasets, demonstrating its ability to capture complex distributional structures while preserving a tractable semi-parametric form. Full article

Accurately characterizing the statistical fluctuations of vehicle radar cross sections (RCSs) across polarization states and azimuthal sectors is essential for evaluating detection performance, conducting probabilistic simulations, and analyzing target features in millimeter-wave radar systems. Existing one-dimensional RCS statistical models, including Weibull, Chi-square, Lognormal, Rice, and Gaussian distributions, are often limited by their restricted functional expressiveness, making it difficult to simultaneously capture skewness, tail thickness, and azimuthal dependence under narrow angular-domain conditions. In addition, purely nonparametric approaches tend to produce spurious modes under finite-sample conditions and lack interpretable structural priors. To address these limitations, this paper proposes a Unimodal RCS Semiparametric Density Estimator (URCS-SDE) tailored for ground-vehicle targets. The proposed approach adopts kernel density estimation (KDE) as a data-driven baseline representation and incorporates physically plausible structural constraints through unimodal shape projection. Then a beta-type tail template is further introduced in the normalized amplitude domain to regulate boundary decay behavior. Finally, weighted least-squares calibration is performed on the histogram grid of the empirical probability density function (PDF), achieving a balanced trade-off between fitting accuracy and stability in both the peak and tail regions. Using multi-azimuth RCS measurements of two representative ground vehicles, the URCS-SDE is systematically compared with five classical parametric distributions and a representative regularized mixture density network (MDN) baseline. Performance is evaluated under both full-azimuth and directional-window conditions using the sum of squared errors (SSE), root mean squared error (RMSE), coefficient of determination (R-square) and held-out negative log-likelihood (NLL). The results show that the URCS-SDE consistently provides the most accurate and stable density estimates, especially in narrow angular windows. In addition, a threshold-based detection-support example derived from the fitted PDFs demonstrates that the advantage of the URCS-SDE transfers from density reconstruction to a directly engineering-relevant downstream quantity. Full article

Nabkhas are a common type of biogenic aeolian landform in arid and semi-arid regions. Their morphological characteristics, surface sediment grain size composition, and spatial distribution patterns can, to some extent, be associated with the interactions between vegetation and the aeolian environment. In this study, nabkhas formed around Caragana tibetica shrubs in the desert steppe of Damao Banner, Inner Mongolia, were selected as the research object. Based on field investigations, UAV image identification, grain size analysis, and spatial point pattern analysis, the characteristics of nabkhas were comparatively analyzed among a control plot without shrubs (CK) and three shrub-covered plots: a low coverage plot (LCP), a medium coverage plot (MCP), and a high coverage plot (HCP). The results showed that (1) some morphological parameters of nabkhas varied among plots with different vegetation cover, but the responses of various indicators were not entirely consistent. The MCP exhibited relatively higher values in indicators such as shrub long axis (L_g), short axis (W_g), and windward slope length (L_y). (2) The surface sediments of nabkhas were mainly composed of silt and fine sand, followed by very fine sand. Compared with the CK, the silt content was generally lower in the shrub-covered plots, whereas the contents of fine sand and very fine sand were higher. The mean grain size (Mz, Φ value) tended to decrease, while the skewness (SK_G) and kurtosis (K_G) tended to increase, and the sorting coefficient (σ_G) showed relatively limited variation. (3) In the LCP, MCP, and HCP, the fractal dimension (D) was significantly positively correlated with the Mz and σ_G (p < 0.05), and significantly negatively correlated with the SK_G and K_G (p < 0.01), suggesting that the D may be associated with variations in sediment grain size structure. (4) Overall, the nabkhas around Caragana tibetica shrubs exhibited a spatial distribution pattern characterized by aggregation at small scales and randomness at large scales, with small-scale clustering being more evident in the MCP and HCP. In general, nabkhas around Caragana tibetica shrubs under different vegetation cover conditions showed observable differences in morphological characteristics, surface sediment grain size composition, and spatial distribution patterns, providing a comparative case reference for the study of nabkhas in desert steppe areas. Full article

Platelet-rich plasma (PRP) therapies are increasingly used in musculoskeletal and regenerative medicine; however, substantial variability in reported outcomes persists even when similar preparation protocols are employed. In quantitative terms, PRP preparation can be interpreted as a stochastic process in which uncertainty propagates through multiple biological and technical inputs. Herein we propose a probabilistic framework to quantify variability in the platelet dose delivered (PDD) using Monte Carlo simulations. The platelet dose was formulated as a random variable defined by a multiplicative model involving the platelet count (modeled as a normal distribution), concentration factor (log-normal), injected volume (uniform), and processing efficiency (beta). Input parameters were represented by probability distributions derived from ranges reported in the literature, and uncertainty propagation was explored through 100,000 Monte Carlo iterations. The resulting simulations revealed a wide dispersion in PDD, characterized by a right-skewed distribution with a median of 3.1 × 10⁹ platelets and an interquartile range of 1.9 × 10⁹ platelets, yielding a coefficient of variation exceeding 50%. Sensitivity analysis based on variance-based global sensitivity measures (Sobol indices) identified the injected volume and concentration factor as the dominant contributors to output variability, with substantial interaction effects between these parameters accounting for a considerable portion of total variance. The baseline platelet count and processing efficiency had comparatively smaller effects. These results demonstrate how probabilistic modeling can clarify the sources of variability in PRP preparation and provide a generalizable framework for uncertainty quantification in multiplicative biomedical systems. Full article

Diabetic cardiomyopathy (DCM) features progressive fibrotic remodeling, but the shared molecular circuitry connecting diabetes mellitus (DM) to cardiomyopathy (CM) remains unclear. We integrated three DM- and three CM-related Gene Expression Omnibus (GEO) datasets and corrected batch effects with sva, verified by violin plots, principal component analysis (PCA), and silhouette coefficients computed on all common genes (DM: 0.9489 to −0.1016; CM: 0.9693 to −0.045; PC1/PC2 inter-batch differences abolished after normalization). Differential expression analysis identified 2562 DM Differentially expressed genes (DEGs) and 1414 CM DEGs, and their intersection yielded 91 common DEGs (51 upregulated, 40 downregulated). Protein–protein interaction (PPI) analysis prioritized 25 hub genes, whose enrichment profiles implicated insulin resistance/insulin signaling and adrenergic signaling in cardiomyocytes. TRRUST-based inference further defined a regulatory network centered on seven key genes (HIF-1α, ACTN4, ABCB1, LTBP1, CLU, TIMP2, and MYH11). To nominate a candidate target of oxymatrine (OMT), we performed docking and molecular dynamics (MD) simulations for representative complexes; OMT showed the most stable interaction with LTBP1, maintaining a consistently short pocket distance (~0.2 nm), the highest contact frequency, and the lowest MM/PBSA binding free energy (−15.32 kcal/mol), with favorable contributions dominated by van der Waals and nonpolar solvation terms. In primary cardiac fibroblasts (CFs), high glucose (HG, 30 mM glucose) induced proliferative and profibrotic activation, whereas OMT (0.4–0.8 mM) reduced HG-driven proliferation without detectable toxicity below 1.2 mM, suppressed FN, collagen I/III, and α-SMA expression, and inhibited migration. OMT also normalized HG-induced cell-cycle skewing by restoring G0/G1-phase occupancy and reducing S-phase entry, with effects comparable to metformin. Finally, HG increased LTBP1 expression and upregulated SMAD3/SMAD4, while OMT attenuated LTBP1 induction and suppressed downstream TGF-β/SMAD activation. Together, these data integrate cross-dataset transcriptomics with mechanistic validation to position LTBP1 as a putative antifibrotic node targeted by OMT, supporting inhibition of the LTBP1/TGF-β/SMAD axis as a candidate strategy to counter DCM-associated fibrosis. Full article

Uneven seed spacing, skewed stalk posture, and inconsistent planting depth remain major challenges in horizontal sugarcane planting. To address these issues, a semi-automatic transverse sugarcane planter integrating a supply–buffer–discharge seeder and multiple soil-engaging components was developed. The seed placement process and the interaction between stalk discharge and soil disturbance were investigated through Discrete Element Method (DEM) simulations and experiments. First, the working principle and key component parameters of the whole machine were determined. It integrated the processes of soil crushing, furrowing, seeding, ridge covering. In addition, a dynamic analysis was conducted on the inter-particle disengagement effect during the two-step seed filling process of lifting and discharging. Secondly, a discrete element simulation model for the entire process of soil-engaging seed arrangement operations was established for the machine. The effects of forward speed and seed outlet position were studied using a discrete element method (DEM) simulation model that coupled soil disturbance flow with stalk-seed discharge behaviour. Furthermore, a response surface methodology (RSM) experiment was performed on the seeding test bench to quantify the effects of guiding parameters on seed placement uniformity. The determination coefficient (R²) of the established regression model exceeded 0.9, indicating high prediction accuracy. The optimal collaborative parameter combination was optimized as follows: forward speed of 1.2 m·s⁻¹, buffer inclination angle of 55°and supply roller speed of 26 r·min⁻¹. After verification, the seed placement uniformity coefficient of the seeder reached 91.8 ± 1.4%, which met the expected accuracy requirements for horizontal planting. Full article

Background/Objectives: This paper investigates the application of radial basis function (RBF) interpolation to adapt the Toyota Human Model for Safety (THUMS) version 6 finite element (FE) models to diverse anthropometric profiles using ANSUR II data. The research focuses on generating personalized human body models (HBMs) across 50th, 80th, and 98th percentiles for both sexes in standing and seated postures, evaluating mesh quality with quantitative metrics, and assessing posture-dependent transformations. Methods: The geometric accuracy for the standing configuration was quantified using DICE similarity coefficients and the 95th percentile Hausdorff distance (HD95). Results: While global whole-body DICE similarity averaged approximately 0.40 due to an inherent variability in distal limb positioning, regional analysis demonstrated strong volumetric overlap in the critical chest and torso regions with DICE values ranging from 0.80 to 0.88. Regional HD95 values were within 20–30 mm across most of the surface area. Surfaces distance analyses showed that more than 95% of the nodes were within $\pm 20$ mm of the target surfaces with the distribution centered near zero across all the percentiles. The mesh quality for both standing and seated morphs demonstrated low violation rates with the aspect ratio being 28% to 30%, while warpage, skewness and, Jacobian determinants were less than 15%. The seated morphs preserved anatomical alignment and posture despite mesh density differences between the postures. Conclusions: These findings indicate that the morphing process preserves anatomical fidelity while highlighting the need for further optimization to mitigate localized distortions in dynamic simulations. Full article

This study develops and evaluates simultaneous confidence interval procedures for all pairwise differences of coefficients of variation under delta-inverse Gaussian distributions. The objective is to provide reliable comparative inference for relative variability in zero-inflated and highly skewed data, where standard normal-based methods may be unreliable. Five approaches were studied and compared in terms of coverage probabilities and average widths: generalized confidence interval, adjusted generalized confidence interval, fiducial confidence interval, method of variance estimates recovery, and normal approximation. A Monte Carlo simulation study was conducted under varying shape parameters, zero-inflation probabilities, sample sizes, and numbers of populations ($k$ = 3, 6, and 10). Although most methods produced CPs near the nominal 0.95 level, meaningful differences emerged when both coverage accuracy and interval efficiency were considered. The AGCI method consistently delivered stable coverage across parameter settings and remained robust as dimensionality increased. The MOVER approach achieved competitive coverage while frequently yielding narrower intervals. In contrast, GCI occasionally showed mild undercoverage, and FCI tended to produce overly wide intervals. An empirical application to zero-inflated mortality data supports the simulation findings. Overall, AGCI and MOVER provide reliable and practical tools for simultaneous inference on differences in CVs across delta-IG populations. Full article

Sustainable urban development increasingly depends on understanding how digital activity is distributed across space and time, yet the spatiotemporal dynamics of the urban digital landscape remain poorly mapped by conventional data sources. This study uses Beijing as an empirical testbed, applying a multi-dimensional analytical framework to massive mobile network traffic data to decode the metabolic rhythms, distributional laws, and functional organization of the urban digital landscape. The results reveal three findings. First, the urban digital landscape exhibits a sleepless trapezoidal temporal rhythm characterized by continuous saturation without a midday trough and a quantifiable weekend activation lag, indicating that digital metabolism is structurally decoupled from physical mobility patterns. Second, digital traffic follows a skew-normal distribution consistent with a 20/70 rule of spatial polarization, in which the top 20% of super-connector nodes sustain approximately 70% of total urban digital flow, yielding a Gini coefficient of 0.68 as a measurable indicator of infrastructure inequality and systemic vulnerability. Third, four distinct functional prototypes are identified—ranging from continuously active metropolitan cores to inverse-tidal ecological peripheries—empirically validating Beijing’s polycentric transformation through the lens of digital flows. These findings demonstrate that large-scale mobile network traffic data offers a replicable and structurally distinct lens for sustainable urban digital governance, supporting resilient network planning, equitable allocation of digital resources, and evidence-based monitoring of urban functional transformation in rapidly growing megacities. Full article

Camellia reticulata is a valuable woody species prized for both its ornamental and oil-producing qualities. This study focused on four qualitative traits and nine quantitative traits of the fruits collected from nine wild populations and 30 cultivated varieties of C. reticulata. Multivariate statistical methods were employed to analyze the variation patterns of these fruits among populations and varieties, aiming to provide a scientific basis for the resource utilization and genetic improvement of this species. The results showed that the pericarp color clustered into two series: an orange-yellow (red) series (found in eight populations and all 30 cultivars) and a yellow-green series (unique to the Heiniu Mountain I population). The a* value was identified as the key indicator for distinguishing between these two color-series. The fruit shape was predominantly spherical, the seed shape was mostly hemispherical, and the seed coat color was primarily brown. Significant differences (p < 0.05) were observed among the nine quantitative phenotypic traits. Fruit weight exhibited the greatest variation (ranging from 28.499 g to 149.068 g), with particularly prominent differences among populations (Fengqing I was the heaviest at 149.068 g, while Yongping I was the lightest at 28.499 g). The coefficients of variation (CV) for phenotypic traits within populations ranged from 17.209% to 60.803% (mean 31.655%), and within varieties from 13.951% to 72.911% (mean 35.290%). Based on CV weights, seed weight showed the largest variation amplitude (21.342%) among populations, while seed number showed the largest variation amplitude (22.956%) among varieties. Correlation analysis revealed that all nine traits exhibited highly significant correlations across different populations and cultivars. Principal component analysis (PCA) indicated that the eigenvalues of the first two principal components were both greater than 1.00, with cumulative contribution rates reaching 73.570% for populations and 76.064% for cultivars, respectively. Cluster analysis grouped the studied materials into three clusters. The comprehensive evaluation identified the cultivar ‘Lichan’ as having the optimal performance (F = 2.410). Box plots revealed greater dispersion in seed number and pericarp thickness within wild populations, while cultivated varieties showed a wider distribution in locule number and fruit transverse diameter. Frequency distribution analysis demonstrated that all traits followed a normal distribution (R² = 0.673~0.999). Among them, fresh seed weight and fruit transverse diameter displayed obvious skewness. Furthermore, the variation in seed number was significantly higher in wild populations than in cultivars. This study reveals rich phenotypic variation in fruit traits between wild populations and cultivated groups of C. reticulata, with fruit size and seed number identified as key traits. These findings provide an important basis for the subsequent selection of hybrid parents and breeding of high-yield, high-oil varieties. Full article