National Taitung University

BackgroundWith no approved pharmacological treatments for non-alcoholic fatty liver disease (NAFLD) in Taiwan, identifying protective and risk factors is crucial for preventing disease progression. Given the clinical heterogeneity of NAFLD, this study aimed to identify clinically meaningful NAFLD phenotypes using electronic medical records (EMRs) and unsupervised clustering, stratify risk across different clusters, identify factors associated with disease progression, and derive a parsimonious set of predictors for high-risk phenotypes.MethodsThis study was a retrospective cohort study conducted in three steps with iterative model training. In step 1, patients diagnosed with NAFLD were identified, and all relevant patient data were extracted, followed by clustering analysis using the k-prototype algorithm. In step 2, survival analysis and Cox regression were applied to perform risk stratification across clusters. In step 3, Lasso regression, logistic regression, and receiver operating characteristic (ROC) curve analysis were used to identify potential protective and risk factors associated with NAFLD and to derive a parsimonious set of predictors for high-risk phenotypes across different risk strata.ResultsStep 1: The analysis of 6,023 patients identified four distinct phenotypic clusters. The first cluster had the most severe disease, the second the least. Step 2: Among 4,998 patients, the first cluster faced the highest risk for all outcomes, with a median survival of 3.06 years, significantly different from the others. There was no significant risk difference between the second and third clusters. Step 3: A comparison of the highest-risk and lowest-risk clusters finally identified 17 potential variables.ConclusionsUsing multiple analytical models, this study identified 17 potential risk factors associated with NAFLD progression. Their combined assessment may inform future risk stratification and hypothesis generation. Further validation is required before clinical application.

Multi-Access Edge Computing (MEC) is a fundamental component for 5G networks to overcome the latency limitations of traditional cloud computing. However, bringing resources closer to users exposes edge nodes to significant security threats, particularly volumetric Denial of Service (DoS) attacks. Current defenses often depend on static thresholds or computationally expensive deep learning, which can exhaust the limited resources of MEC nodes. To address these limitations, this paper proposes a resource-optimized edge-centric security management logic that integrates Software Defined Network (SDN) with lightweight supervised learning (C5.0, Bagging-CART, and Random Forest). Unlike standard system integrations, we introduce a dynamic non-permanent blocking algorithm designed to balance detection accuracy with control plane stability. Experimental results demonstrate that the proposed C5.0 model, operating at a specific 0.20% sFlow sampling point, achieves 100% detection accuracy with under 100 ms mitigation latency. The system successfully reduces volumetric attack loads from 445 Mbps to 95 Mbps (a 78% reduction) at the node level. These findings confirm that the proposed framework achieves higher computational efficiency than complex alternatives, making it a highly stable solution for constrained 5G MEC environments.

Accurate identification of process parameters from in-mold sensing signals is essential for advancing intelligent and adaptive injection molding systems. This study proposes a data-driven inverse modeling framework that infers key process parameters directly from cavity pressure curves. A stage-wise Autoencoder (AE) is employed to perform nonlinear feature extraction by segmenting the pressure curve into filling, packing, and cooling stages, thereby capturing process-dominant dynamics. The encoded latent features are subsequently mapped to injection speed, packing pressure, nozzle temperature, and mold temperature using a multilayer perceptron (MLP). Eighty-one full-factorial parameter combinations were experimentally conducted, generating 810 molding cycles on an industrial injection molding machine. An independent validation dataset was collected from separate production runs to evaluate generalization capability. The proposed AE–MLP framework achieved high predictive accuracy on unseen data, with R² values of 0.9979 for injection speed, 0.9926 for packing pressure, 0.9571 for mold temperature, and 0.9296 for nozzle temperature. The corresponding RMSE values remained below 1.6 across all parameters. Comparative analysis against PCA–MLP and direct MLP models demonstrates that nonlinear manifold representation significantly improves inverse regression robustness, particularly for thermally coupled parameters. Furthermore, the stability of inverse modeling is analyzed from the perspective of parameter identifiability and many-to-one mapping characteristics inherent in injection molding processes. Results indicate that stage-wise nonlinear feature encoding enhances separability in the latent space and mitigates ambiguity in parameter inference. The proposed framework provides a robust and physically interpretable approach for real-time process parameter estimation, contributing to the development of adaptive and self-optimizing manufacturing systems. Infers molding parameters directly from multi-sensor pressure features. AE–MLP framework learns nonlinear relations between pressure and parameters. Parameter recommendation enables adaptive process condition adjustment. 810-cycle full-factorial experiments confirm accuracy and generalization. Supports smart manufacturing through real-time parameter tuning.

Low birth weight (LBW) is a leading cause of death for newborns and increases chronic disease risks later in life. Early identification of LBW risk is crucial. The objective of this study was to develop predictive models for LBW using boosting ensemble machine learning, with a focus on features available during early pregnancy, such as pre-pregnancy body mass index, body height, and blood pressure before 20 weeks of pregnancy. This is a retrospective cohort study. We used electronic medical records in four hospitals in Taiwan where pregnant women received prenatal care from January 2016 to July 2019, including 6719 pregnant women. Data preprocessing involved normalization, one-hot encoding, and a synthetic minority oversampling technique for class imbalance. Boosting ensemble methods were used to build the LBW predictive models. The mean diastolic blood pressure (DBP) in early pregnancy (<20 weeks) was 66.5 mmHg, 29.6% had experienced abortion, 8.7% delivered LBW, 12.2% were overweight or obese before pregnancy, and 18.3% had elevated or stage I hypertension before 20 weeks of pregnancy. Lightweight Gradient Boosting Machine was the best-performing LBW model, with an area under curve of 0.96 and an accuracy of 93.4%. Early pregnancy DBP, maternal height, and number of abortions were the most important features. The LBW prediction model performed well. Nurses could use the model to assess LBW risk and intervene early. Preventive efforts could be directed to blood pressure management starting early pregnancy, nutritional support for short mothers, and self-care for women with a history of abortions.

Abstract By using graphene as the carrier transport layer and transition metal dichalcogenides (TMDs, WS 2 , MoS 2 , and WSe 2 ) as the light absorption layer, wavelength-tunable photodetectors can be fabricated on sapphire substrates. With the short carrier transport time in the graphene layer, the hole accumulation in the TMD layers would result in the conduction change in the graphene layer. Therefore, high responsivity values of 1521.9, 6077.9 and 3977.8 A/W at 630, 660 and 750 nm are observed for the WS 2 , MoS 2 and WSe 2 devices, respectively. With the high responsivity values, high detectivity values 6.9 (WS 2 ), 17.6 (MoS 2 ) and 9.2 (WSe 2 ) × 10 10 Jones are also observed for the three devices. Following the similar film transferring procedure and device fabrication procedure, photodetectors with mono-layer graphene on different mono-layer TMDs are fabricated on polyethylene terephthalate (PET) substrates. High responsivity values (10 2 -10 3 A/W) are still observed for the devices fabricated on flexible substrates. Although a relatively larger responsivity reduction is observed for the graphene/MoS 2 device under bending conditions, the similar responsivity values observed for the other two devices (graphene/WS 2 and graphene/WSe 2 ) still indicate that the thin-body nature of 2D materials is advantageous for the fabrication of flexible devices. With increasing WSe 2 layer numbers, even higher responsivities and significantly shortened rise/fall times 4.3/13.0 ms are observed for the mono-layer graphene/tri-layer WSe 2 photodetector, which has demonstrated that the excess electron storage in multi-layer 2D materials would result in fast hole accumulation/relaxation in the WSe 2 light absorption layer. The easy replacement of 2D materials with different bandgap values for wavelength-tunable detections also indicate an alternate application other than electronic devices of 2D materials in weak light detections.

Certification-based vocational education often emphasizes practical training, yet repetitive subject-based learning tasks place a heavy burden on teachers and reduce student motivation. To address this challenge, this study examined the integration of a gamified chatbot into cognitive content instruction for the Level B Computer Hardware Fabrication certification. A quasi-experimental design was implemented across three academic years, involving a control group (conventional instruction), an experimental group using a non-gamified chatbot, and another using a gamified chatbot. Over an eight-week intervention, participants completed pre- and post-tests to measure learning effectiveness, while learner satisfaction was assessed through a validated questionnaire. Results from ANCOVA revealed that the gamified chatbot group achieved learning outcomes equivalent to teacher-led instruction and significantly outperformed the non-gamified chatbot group. Post hoc tests confirmed large effect sizes favoring gamification. Learners also reported greater satisfaction, particularly in reduced boredom and improved alignment with learning preferences. These findings demonstrate that a gamified chatbot can effectively function as a mobile cognitive content instructor, sustaining motivation, enhancing learning outcomes, and alleviating teacher workload in certification-oriented education. The proposed model is scalable and holds global relevance, offering adaptability to other vocational certifications, STEM training, and content-intensive learning contexts.