Predicting student performance is essential for educational institutions to identify at-risk students and offer necessary support. However, traditional models often struggle with class imbalance, resulting in biased performance, particularly for minority classes. In this article, a novel class-based multilayer CatBoost is proposed for student performance prediction, in which each layer focuses on a binary classification task. At each layer, the training set is divided into two groups, one is the majority class, and the other one is the remaining classes, thereby improving class balance. The majority class samples are then excluded, and the remaining samples are used to construct the next layer. This procedure is repeated until two classes remain for training. During prediction, a layerwise exact matching mechanism is implemented. Starting from the first layer, if a sample is classified as the target class, the corresponding label is returned as the final prediction; otherwise, the sample is passed to the next layer, and the matching procedure is repeated. Experimental results demonstrate that the proposed model outperforms other comparison models, highlighting the effectiveness of the class-based multilayer structure in handling class imbalance. This suggests that the proposed model could support data-driven decision-making and contribute to improving educational quality.
Contribution: This study presents flip-game engineering and technology (Flip-GET), a blended instructional methodology that integrates serious games (SGs) with flipped classroom (FC) principles to support lab-based engineering education. Unlike some traditional formats, Flip-GET proposes a scalable model designed to enhance student preparation, performance, and engagement, while potentially optimizing faculty workload. Background: Engineering education fundamentally relies on laboratory instruction. However, increasing enrollment, limited infrastructure, and the necessity for repetitive explanations have challenged the sustainability of traditional methods. Flip-GET seeks to address these issues by enabling students to rehearse experimental procedures digitally prior to physical lab sessions. Intended outcomes: The model aims to: 1)improve academic performance; 2)reduce procedural errors; 3)enhance student motivation and confidence; 4)minimize instructional repetition; and 5) increase overall efficiency in lab instruction. Application design: Flip-GET employs gamified simulations as prelab activities to prepare students for in-person sessions. This instructional design is grounded in active learning theory and design-based research (DBR) principles. Findings: Data collected over three academicyears from 352students indicate that Flip-GET was associated with a 24% increase in posttest scores, a 37% reduction in procedural errors, and a 25% improvement in analytical reasoning. Furthermore, observed lab time decreased by 67% and instructional repetition by 60%. A93% of participants reported greater understanding and motivation, suggesting that Flip-GET holds significant potential for optimizing physical infrastructure and diversifying the practical curriculum in engineering education.
Contribution: A Class–Lab–Field (CLF) framework for structural health monitoring (SHM) education centers on ArduinoNode, an open, ultralow-cost (~U.S. ${\$}50$ /node) wireless Internet of Things (IoT) platform students build, program, and deploy, narrowing the analysis–field gap from sensing to interpretation. Background: SHM courses foreground structural models and algorithms over instrument working principles and operational procedures; many Arduino labs lack coordinated outdoor deployment. Inspectable, affordable kits for teaching instrument internals remain scarce. Intended outcomes: Operational literacy along the SHM pipeline: accelerometer fundamentals, instrument limits, and links among acquisition, synchronization, logging, and defensible interpretation of noisy field data. Application design: CLF sequences classroom node preparation, lab work on a structural model, and field multisystem monitoring for compressed programs. Hands-on sensing comes first; SHM theory depth scales with contact hours. Findings: CLF: APESS2025 ( $N=56$ ): 93.3% satisfaction, 100% recommend; ~60% report sensor/SHM/IoT gains, 40% WSN gains, and 60% rate hands-on emphasis much or significantly above typical theory-heavy courses. ArduinoNode: Footbridge data (ArduinoNode, smartphones, commercial SHM) compared with a finite-element (FE) model yield 5.1% and 7.2% errors on the first two modes under dual-link, sub-10-ms-class synchronization.
Contribution: Despite their logistical advantages, multiple-choice questions (MCQ) and multiple-true–false (MTF) formats are often underused or viewed skeptically in science, technology, engineering, and mathematics (STEM) education due to concerns about superficiality, a lack of depth, and vulnerability to guessing. This study presents redesigned MCQ and MTF formats that aim to improve objectivity, fairness, and efficiency while preserving conceptual rigor. Background: Open-ended questions are traditionally favored in STEM due to their ability to assess reasoning and partial understanding. However, they pose challenges for consistent and efficient grading, especially in large courses. Meanwhile, the literature shows that well-constructed objective items can match open-ended questions in diagnostic power, yet most existing MCQs fail to meet quality standards, and MTFs remain underutilized. Intended outcomes: The intervention seeks to: 1) ensure objective, deterministic grading; 2) reduce the impact of guessing; 3) support partial credit where appropriate; and 4) offer an efficient solution without compromising the depth of assessment. Application design: We propose a ten-option MCQ structure designed to reduce random success rates while enabling richer distractor design, and a sum-based MTF variant where students report the sum of weights assigned to true statements. The combination of both formats enables automatic grading, discourages guessing, and allows nuanced performance measurement. These were applied to two exams in an undergraduate Electrical Circuits II course over a semester. Findings: Results from 52 students indicate that the proposed formats produce balanced precision and effective discrimination between items. Partial credit allowed for better scoring without inflating the results. A parallel evaluation using state-of-the-art large-language models revealed low performance on the proposed questions, suggesting that such structured formats cannot yet be reliably replaced by AI-based grading of open-ended STEM responses. These findings suggest that thoughtfully redesigned objective questions can serve as efficient, fair, and pedagogically rigorous alternatives to open-ended formats in theoretical or foundational STEM courses where conceptual precision is paramount, and should be complemented by project-based or laboratory assessments.
