Training Data Research Articles

Using ADTS and CNN methods to effectively monitor CD, crosstalk, and OSNR in an optical network

The article presents the results of a method based on asynchronous delay-tap sampling (ADTS) and convolutional neural network (CNN) for determining simultaneously occurring disturbances described using the chromatic dispersion (CD), crosstalk and optical signal-to-noise ratio (OSNR) parameters. The ADTS method was used to generate training and test data for the convolutional network, which in turn was used to learn to recognize interference from said data. The tests were carried out for a transmission speed of 10 Gbit/s and for on-off keying (OOK) and differential phase shift keying (DPSK) modulation. Very good results were obtained in recognizing simultaneously occurring phenomena. Accuracy of over 99% was achieved for CD and crosstalk for DPSK modulation and over 98% for OOK modulation. In the case of amplified spontaneous emission (ASE) noise, slightly weaker results were obtained, above 95–96% for both modulations. Based on the conducted research, it was determined that the use of ADTS and CNN methods enables monitoring of simultaneously occurring CD, crosstalk, and ASE noise in the physical layer of the optical network, while maintaining the requirements for modern monitoring systems.

Developing large language models to detect adverse drug events in posts on x

ABSTRACT Adverse drug events (ADEs) are one of the major causes of hospital admissions and are associated with increased morbidity and mortality. Post-marketing ADE identification is one of the most important phases of drug safety surveillance. Traditionally, data sources for post-marketing surveillance mainly come from spontaneous reporting system such as the Food and Drug Administration Adverse Event Reporting System (FAERS). Social media data such as posts on X (formerly Twitter) contain rich patient and medication information and could potentially accelerate drug surveillance research. However, ADE information in social media data is usually locked in the text, making it difficult to be employed by traditional statistical approaches. In recent years, large language models (LLMs) have shown promise in many natural language processing tasks. In this study, we developed several LLMs to perform ADE classification on X data. We fine-tuned various LLMs including BERT-base, Bio_ClinicalBERT, RoBERTa, and RoBERTa-large. We also experimented ChatGPT few-shot prompting and ChatGPT fine-tuned on the whole training data. We then evaluated the model performance based on sensitivity, specificity, negative predictive value, positive predictive value, accuracy, F1-measure, and area under the ROC curve. Our results showed that RoBERTa-large achieved the best F1-measure (0.8) among all models followed by ChatGPT fine-tuned model with F1-measure of 0.75. Our feature importance analysis based on 1200 random samples and RoBERTa-Large showed the most important features are as follows: “withdrawals”/”withdrawal”, “dry”, “dealing”, “mouth”, and “paralysis”. The good model performance and clinically relevant features show the potential of LLMs in augmenting ADE detection for post-marketing drug safety surveillance.

Implementasi Data Mining Klasifikasi Gejala Penyakit TB Menggunakan Algoritma Naive Bayes pada Studi Kasus Puskesmas Pegangsaan Dua B

Tuberculosis (TB) is an infectious disease caused by Mycobacterium tuberculosis. The bacteria can enter the lungs, causing the patient to experience shortness of breath and chronic coughing. The infection process in TB progresses through three main stages: primary infection, latent infection, and active infection. During the primary infection stage, the bacteria enter the lungs through inhaled air. If the immune system successfully controls the bacteria, the infection becomes latent, where the bacteria remain inactive and cause no symptoms. However, if the immune system fails, the infection becomes active, and the bacteria attack healthy lung cells, leading to severe symptoms. Tuberculosis remains a major cause of global morbidity and mortality. According to the 2022 WHO report, Indonesia ranks second in the world with an estimated 969,000 TB cases, or 354 per 100,000 people, and a mortality rate of 144,000, or 52 per 100,000 people. This study aims to improve the accuracy of TB diagnosis through the application of the Naive Bayes algorithm, applied to patient data from the Pegangsaan Dua B Community Health Center. The dataset includes 1,527 training data and 1,029 testing data, covering attributes such as patient name, gender, age, identification number, registration date, address, diagnosis status, and diagnosis results. The results indicate that the Naive Bayes algorithm achieved an accuracy of 99.71%, showing its potential for early and accurate TB diagnosis prediction.

Deep learning–based statistical robustness evaluation of intensity-modulated proton therapy for head and neck cancer

A dense, dilated 3D U-net was trained to predict the 5th and 95th percentile distributions of planning dose plan using the nominal dose and planning CT images. The data set comprised proton therapy plans for 582 H&N cancer patients. Ground truth percentile values were estimated for each patient through 600 dose recalculations, representing randomly sampled uncertainty scenarios. The comprehensive comparisons of different models were conducted for H&N cancer patients, considering those with and without a beam mask (BM) and diverse beam configurations, including varying beam angles, couch angles, and beam numbers. The performance of our model trained based on a mixture of patients with H&N and prostate cancer was also assessed in contrast with models trained based on data specific for patients with cancer at either site. The DL-based model's predictions of percentile dose distributions exhibited excellent agreement with the ground truth dose distributions. The average gamma index with 2 mm/2%, consistently exceeded 97% for both 5th and 95th percentile dose volumes. Mean dose-volume histogram error analysis revealed that predictions from the combined training set yielded mean errors and standard deviations that were generally similar to those in the specific patient training data sets. Our proposed DL-based method for evaluation of the robustness of proton therapy plans provides precise, rapid predictions of percentile dose for a given confidence level regardless of the beam arrangement and cancer site. This versatility positions our model as a valuable tool for evaluating the robustness of proton therapy across various cancer sites.

Prediction of preterm birth in multiparous women using logistic regression and machine learning approaches

To predict preterm birth (PTB) in multiparous women, comparing machine learning approaches with traditional logistic regression. A population-based cohort study was conducted using data from the Ontario Better Outcomes Registry and Network (BORN). The cohort included all multiparous women who delivered a singleton birth at 20–42 weeks’ gestation in an Ontario hospital between April 1, 2012 and March 31, 2014. The primary outcome was PTB < 37 weeks, with spontaneous PTB analyzed as a secondary outcome. Stepwise logistic regression and the Boruta machine learning were used to select the important variables during the first and second trimester. For building prediction models, the whole data set were divided for the two independent parts: two-third for training the classifiers (Logistic regression, random forests, decision trees, and artificial neural networks) and one-third for model validation. Then, the training data set were balanced by random over sampling technique. The best hyper parameters were obtained by the tenfold cross validation. The performance of all models was evaluated by sensitivity, specificity, positive predictive value, negative predictive value, and the area under the receiver operating characteristics (AUC). The cohort included 145,846 births, of which 8125 (5.57%) were preterm. In first-trimester models, the strongest predictors of PTB were previous PTB, preexisting diabetes, and abnormal pregnancy‐associated plasma protein-A. In the testing data set, the highest predictive ability was seen for artificial neural networks, with an area under the receiver operating characteristic curve (AUC) of 68.8% (95% CI 67.6–70.1%). In second-trimester models, addition of infant sex, attendance at first-trimester appointment, medication exposure, and abnormal alpha-fetoprotein concentrations increased the AUC to 72.1% (95% CI 71.1–73.1%) with logistic regression. With the inclusion of the variable complications during pregnancy, the AUC increased to 80.5% (95% CI 79.6–81.5%) using logistic regression. For both overall and spontaneous PTB, during both the first and second trimesters, models yielded negative predictive values of 97%. Overall, machine learning and logistic regression produced similar performance for prediction of PTB. For overall and spontaneous PTB, both first- and second-trimester models provided negative predictive values of ~ 97%, higher than that of fetal fibronectin.

Fintech model performance analysis in credit loans: evidence from China city commercial banks

ABSTRACT The analysis of bank loan performance has consistently been a central issue in banking research. With the impact of the COVID-19 pandemic, we have witnessed a significant increase in online loans. Our study employs fintech models to develop a big data credit analysis system, evaluating model efficiency and the effect of ‘data organization’ on predictive performance for city commercial bank loans. Our results show that training data size, timeliness, and inter-bank ‘data sharing’ significantly affect model predictions. We also find that traditional econometric models (Logit and Probit) have the weakest predictive ability. However, ‘most advanced’ deep learning models perform worse than machine learning models with lower algorithmic complexity, such as Random Forest. Besides, ‘sharing training data’ generally increases model efficiency, but there are exceptions for larger city commercial banks. This paper provides ample empirical evidence for the application of the fintech model in bank loan credit. It highlights several characteristics of training data, such as size and timeliness, which greatly affect model efficiencies. We also provide policy recommendations for fintech transition and collaboration among municipal banks.

Large language models and humans converge in judging public figures’ personalities

Abstract ChatGPT-4 and 600 human raters evaluated 226 public figures’ personalities using the Ten-Item Personality Inventory. The correlation between ChatGPT-4 and aggregate human ratings ranged from r=.76 to .87, outperforming the models specifically trained to make such predictions. Notably, the model was not provided with any training data or feedback on its performance. We discuss the potential explanations and practical implications of ChatGPT-4’s ability to mimic human responses accurately.

Self-Supervised Learning across the Spectrum

Satellite image time series (SITS) segmentation is crucial for many applications, like environmental monitoring, land cover mapping, and agricultural crop type classification. However, training models for SITS segmentation remains a challenging task due to the lack of abundant training data, which requires fine-grained annotation. We propose S4, a new self-supervised pretraining approach that significantly reduces the requirement for labeled training data by utilizing two key insights of satellite imagery: (a) Satellites capture images in different parts of the spectrum, such as radio frequencies and visible frequencies. (b) Satellite imagery is geo-registered, allowing for fine-grained spatial alignment. We use these insights to formulate pretraining tasks in S4. To the best of our knowledge, S4 is the first multimodal and temporal approach for SITS segmentation. S4’s novelty stems from leveraging multiple properties required for SITS self-supervision: (1) multiple modalities, (2) temporal information, and (3) pixel-level feature extraction. We also curate m2s2-SITS, a large-scale dataset of unlabeled, spatially aligned, multimodal, and geographic-specific SITS that serves as representative pretraining data for S4. Finally, we evaluate S4 on multiple SITS segmentation datasets and demonstrate its efficacy against competing baselines while using limited labeled data. Through a series of extensive comparisons and ablation studies, we demonstrate S4’s ability as an effective feature extractor for downstream semantic segmentation.

RODIN: Raw Diffraction Data for Teaching, Training, and Demonstration

<i>RODIN</i>: Raw Diffraction Data for Teaching, Training, and Demonstration

Machine Learning for Advanced Emission Monitoring and Reduction Strategies in Fossil Fuel Power Plants

Fossil fuel power plants are a significant contributor to global carbon dioxide (CO2) and nitrogen oxide (NOx) emissions. Accurate monitoring and effective reduction of these emissions are crucial for mitigating climate change. This systematic review examines the current state of research on the application of machine learning techniques in evaluating the emissions from fossil fuel power plants. This review first briefly introduces the continuous emission monitoring (CEM) systems and predictive emission monitoring (PEM) systems that are commonly used in power plants and highlights that machine learning models can significantly improve PEM systems through their capability to process and interpret large datasets intelligently to transform traditional emission monitoring systems by enhancing their precision, effectiveness, and cost-efficiency. Compared to previously published review articles, the key contribution and innovation in this present review is the discussion of machine learning models in CO2/NOx emissions according to the different algorithms used, including their advantages and disadvantages in a systematic way, which aims to help future researchers to develop more effective machine learning models. The most popular machine learning model includes reinforcement learning, a forward neural network, a long short-term memory neural network, and support vector regression. While each model method has its own advantages and disadvantages, we noted that training data quality, as well as the proper selection of model parameters, plays an important role. The challenges and research gaps, such as model transferability, a deep understanding of the physics of CO2/NOx emissions, and the availability of high-quality data for training machine learning models, are identified, and recommendations as well as potential future research directions to address these challenges are proposed and discussed.

An Interval Neural Network Method for Identifying Static Concentrated Loads in a Population of Structures

During the design and validation of structural engineering, the focus is on a population of similar structures, not just one. These structures face uncertainties from external environments and internal configurations, causing variability in responses under the same load. Identifying the real load from these dispersed responses is a significant challenge. This paper proposes an interval neural network (INN) method for identifying static concentrated loads, where the network parameters are internalized to create a new INN architecture. Additionally, the paper introduces an improved interval prediction quality loss function indicator named coverage and mean square criterion (CMSC), which balances the interval coverage rate and interval width of the identified load, ensuring that the median of the recognition interval is closer to the real load. The efficiency of the proposed method is assessed through three examples and validated through comparative research against other loss functions. Our research findings indicate that this approach enhances the interval accuracy, robustness, and generalization of load identification. This improvement is evident even when faced with challenges such as limited training data and significant noise interference.

Heterogeneous Transfer Learning of Electrohydrodynamic Printing Under Zero-Gravity Toward In-Space Manufacturing

Abstract As we continue to commercialize space and mature in-space manufacturing (ISM) processes, there is a strong need to transfer the knowledge we learn from experiments on the ground to zero-gravity environments. Physics-motivated manufacturing processes, like additive manufacturing, experience a shift in fabrication parameters due to the absence of gravity and the change of environments. Thus, we found traditional machine learning methods are not capable of addressing this domain shift and present a transfer learning scheme as a solution in this paper. We tested a kernel ridge regression model built for heterogeneous transfer learning (KRR-HeITL) on data from the electrohydrodynamic inkjet printing (EHD printing) process. EHD printing is a process that uses electrical force to control material flows, thus achieving the fabrication of electronics without requiring gravity. Our team has successfully conducted three rounds of parabolic flights to validate this technology for ISM. We trained on multiple datasets built from on-ground experiments and tested using zero-gravity printing data obtained from parabolic flight tests. Measurements of the Taylor cone both on-ground and in zero-gravity were taken and exploited as a part of the training data. We found that our method obtains good interpolation accuracy (MAPE 3.85%) compared to traditional machine learning methods (MAPE 16.84%) for predicting the printed line width. We concluded that the KRR-HeITL method is well suited for zero-gravity domain shifts of EHD printing parameters. This study paves the way for future predictions of ISM parameters when there are only on-ground experiments or very limited zero-gravity datasets for a given process.

A deep learning dataset for metal multiaxial fatigue life prediction

Multiaxial fatigue failure of metals, a common issue in industrial production, often leads to significant losses. Recently, many researchers have applied deep learning methods to predict the multiaxial fatigue life of metals, achieving promising results. Due to the high costs of fatigue testing, training data for deep learning is scarce and labor-intensive to collect. This study meets this need by creating a large-scale, high-quality dataset for multiaxial fatigue life prediction, consisting of 1167 samples from 40 materials collected from literature. The dataset includes key mechanical properties (elastic modulus, yield strength, tensile strength, Poisson’s ratio) and 48 loading paths, along with additional relevant information (composition ratios, processing conditions). Common deep learning models validated the dataset’s effectiveness. This dataset aims to support researchers applying deep learning to fatigue life prediction, addressing the long-standing issue of data scarcity, thereby advancing the intersection of artificial intelligence and metal fatigue research.

SeqSeg: Learning Local Segments for Automatic Vascular Model Construction.

Computational modeling of cardiovascular function has become a critical part of diagnosing, treating and understanding cardiovascular disease. Most strategies involve constructing anatomically accurate computer models of cardiovascular structures, which is a multistep, time-consuming process. To improve the model generation process, we herein present SeqSeg (sequential segmentation): a novel deep learning-based automatic tracing and segmentation algorithm for constructing image-based vascular models. SeqSeg leverages local U-Net-based inference to sequentially segment vascular structures from medical image volumes. We tested SeqSeg on CT and MR images of aortic and aortofemoral models and compared the predictions to those of benchmark 2D and 3D global nnU-Net models, which have previously shown excellent accuracy for medical image segmentation. We demonstrate that SeqSeg is able to segment more complete vasculature and is able to generalize to vascular structures not annotated in the training data.

Online transaction fraud detection in the banking sector using machine learning techniques

In this paper Nowadays, online transactions are increasing rapidly and payment scams are also increasing. A hacker is a person who gains access to another person's financial information. Improving feature selection is important to detect transaction risk in large, multidimensional data. Fraud detection and classification performance is divided into two parts: training data and test data. During this process we have focused on data analysis and pre-processing. In a transaction’s dataset, the predictive variables influence how well a machine learning algorithm for transaction risk identification classifies and identifies fraudulent transactions. This study examines the application of machine learning techniques in improving fraud detection in the banking industry. Using advanced algorithms and historical transaction data, the model aims to identify anomalous patterns that indicate fraudulent activity. Feature engineering, ensemble methods, and anomaly detection algorithms are used to improve the accuracy and performance of fraud detection systems. reducing false positives and provides an adaptive security mechanism. By harnessing the power of machine learning, this approach aims to strengthen the banking sectors defines against sophisticated fraud schemes. This study not only contributes to the academic understanding of fraud detection, but also provides industry practitioners with a practical framework for implementing proactive measures. In navigating the complex landscape of financial security, this research serves as a guide for building agile and responsive systems that stay ahead of the threats that evolve in the ever-changing landscape of the banking industry.

The Evaluation of Machine Learning Techniques for Isotope Identification Contextualized by Training and Testing Spectral Similarity

Precise gamma-ray spectral analysis is crucial in high-stakes applications, such as nuclear security. Research efforts toward implementing machine learning (ML) approaches for accurate analysis are limited by the resemblance of the training data to the testing scenarios. The underlying spectral shape of synthetic data may not perfectly reflect measured configurations, and measurement campaigns may be limited by resource constraints. Consequently, ML algorithms for isotope identification must maintain accurate classification performance under domain shifts between the training and testing data. To this end, four different classifiers (Ridge, Random Forest, Extreme Gradient Boosting, and Multilayer Perceptron) were trained on the same dataset and evaluated on twelve other datasets with varying standoff distances, shielding, and background configurations. A tailored statistical approach was introduced to quantify the similarity between the training and testing configurations, which was then related to the predictive performance. Wilcoxon signed-rank tests revealed that the OVR-wrapped XGB significantly outperformed the other algorithms, with confidence levels of 99.0% or above for the 133Ba, 60Co, 137Cs, and 152Eu sources. The findings from this work are significant as they outline techniques to promote the development of robust ML-based approaches for isotope identification.

Estimation of Source Range and Location Using Ship-Radiated Noise Measured by Two Vertical Line Arrays with a Feed-Forward Neural Network

Machine learning-based source range estimation is a promising method for enhancing the performance of tracking both the dynamic and static positions of targets in the underwater acoustic environment using extensive training data. This study constructed a machine learning model for source range estimation using ship-radiated noise recorded by two vertical line arrays (VLAs) during the Shallow-water Acoustic Variability Experiment (SAVEX-15), employing the Sample Covariance Matrix (SCM) and the Generalized Cross Correlation (GCC) as input features. A feed-forward neural network (FNN) was used to train the model on the acoustic characteristics of the source at various distances, and the range estimation results indicated that the SCM outperformed the GCC with lower error rates. Additionally, array tilt correction using the array invariant-based method improved range estimation accuracy. The impact of the training data composition corresponding to the bottom depth variation between the source and receivers on range estimation performance was also discussed. Furthermore, the estimated ranges from the two VLA locations were applied to localization using trilateration. Our results confirm that the SCM is the more appropriate feature for the FNN-based source range estimation model compared with the GCC and imply that ocean environment variability should be considered in developing a general-purpose machine learning model for underwater acoustics.

A multimodal approach to support teacher, researcher and AI collaboration in STEM+C learning environments

AbstractRecent advances in generative artificial intelligence (AI) and multimodal learning analytics (MMLA) have allowed for new and creative ways of leveraging AI to support K12 students' collaborative learning in STEM+C domains. To date, there is little evidence of AI methods supporting students' collaboration in complex, open‐ended environments. AI systems are known to underperform humans in (1) interpreting students' emotions in learning contexts, (2) grasping the nuances of social interactions and (3) understanding domain‐specific information that was not well‐represented in the training data. As such, combined human and AI (ie, hybrid) approaches are needed to overcome the current limitations of AI systems. In this paper, we take a first step towards investigating how a human‐AI collaboration between teachers and researchers using an AI‐generated multimodal timeline can guide and support teachers' feedback while addressing students' STEM+C difficulties as they work collaboratively to build computational models and solve problems. In doing so, we present a framework characterizing the human component of our human‐AI partnership as a collaboration between teachers and researchers. To evaluate our approach, we present our timeline to a high school teacher and discuss the key insights gleaned from our discussions. Our case study analysis reveals the effectiveness of an iterative approach to using human‐AI collaboration to address students' STEM+C challenges: the teacher can use the AI‐generated timeline to guide formative feedback for students, and the researchers can leverage the teacher's feedback to help improve the multimodal timeline. Additionally, we characterize our findings with respect to two events of interest to the teacher: (1) when the students cross a difficulty threshold, and (2) the point of intervention, that is, when the teacher (or system) should intervene to provide effective feedback. It is important to note that the teacher explained that there should be a lag between (1) and (2) to give students a chance to resolve their own difficulties. Typically, such a lag is not implemented in computer‐based learning environments that provide feedback. Practitioner notesWhat is already known about this topic Collaborative, open‐ended learning environments enhance students' STEM+C conceptual understanding and practice, but they introduce additional complexities when students learn concepts spanning multiple domains. Recent advances in generative AI and MMLA allow for integrating multiple datastreams to derive holistic views of students' states, which can support more informed feedback mechanisms to address students' difficulties in complex STEM+C environments. Hybrid human‐AI approaches can help address collaborating students' STEM+C difficulties by combining the domain knowledge, emotional intelligence and social awareness of human experts with the general knowledge and efficiency of AI. What this paper adds We extend a previous human‐AI collaboration framework using a hybrid intelligence approach to characterize the human component of the partnership as a researcher‐teacher partnership and present our approach as a teacher‐researcher‐AI collaboration. We adapt an AI‐generated multimodal timeline to actualize our human‐AI collaboration by pairing the timeline with videos of students encountering difficulties, engaging in active discussions with a high school teacher while watching the videos to discern the timeline's utility in the classroom. From our discussions with the teacher, we define two types of inflection points to address students' STEM+C difficulties—the difficulty threshold and the intervention point—and discuss how the feedback latency interval separating them can inform educator interventions. We discuss two ways in which our teacher‐researcher‐AI collaboration can help teachers support students encountering STEM+C difficulties: (1) teachers using the multimodal timeline to guide feedback for students, and (2) researchers using teachers' input to iteratively refine the multimodal timeline. Implications for practice and/or policy Our case study suggests that timeline gaps (ie, disengaged behaviour identified by off‐screen students, pauses in discourse and lulls in environment actions) are particularly important for identifying inflection points and formulating formative feedback. Human‐AI collaboration exists on a dynamic spectrum and requires varying degrees of human control and AI automation depending on the context of the learning task and students' work in the environment. Our analysis of this human‐AI collaboration using a multimodal timeline can be extended in the future to support students and teachers in additional ways, for example, designing pedagogical agents that interact directly with students, developing intervention and reflection tools for teachers, helping teachers craft daily lesson plans and aiding teachers and administrators in designing curricula.

AutoPeptideML: A study on how to build more trustworthy peptide bioactivity predictors.

Automated machine learning (AutoML) solutions can bridge the gap between new computational advances and their real-world applications by enabling experimental scientists to build their own custom models. We examine different steps in the development life-cycle of peptide bioactivity binary predictors and identify key steps where automation can not only result in a more accessible method, but also more robust and interpretable evaluation leading to more trustworthy models. We present a new automated method for drawing negative peptides that achieves better balance between specificity and generalisation than current alternatives. We study the effect of homology-based partitioning for generating the training and testing data subsets and demonstrate that model performance is overestimated when no such homology correction is used, which indicates that prior studies may have overestimated their performance when applied to new peptide sequences. We also conduct a systematic analysis of different protein language models as peptide representation methods and find that they can serve as better descriptors than a naive alternative, but that there is no significant difference across models with different sizes or algorithms. Finally, we demonstrate that an ensemble of optimised traditional machine learning algorithms can compete with more complex neural network models, while being more computationally efficient. We integrate these findings into AutoPeptideML, an easy-to-use AutoML tool to allow researchers without a computational background to build new predictive models for peptide bioactivity in a matter of minutes. Source code, documentation, and data are available at https://github.com/IBM/AutoPeptideML and a dedicated web-server at http://peptide.ucd.ie/AutoPeptideML. Raul.fernandezdiaz@ucdconnect.ie or denis.shields@ucd.ie. Supplementary Information can be accessed from Zenodo: Https://zenodo.org/records/13363975.

Transient NOx emission modeling of a hydrogen-diesel engine using hybrid machine learning methods

One promising approach to reduce carbon foot print of internal combustion engines (ICEs) is using alternative fuels like hydrogen, particularly by converting medium and heavy-duty diesel engines to dual-fuel hydrogen-diesel engines. To minimize elevated NOx emissions from hydrogen-fueled engine, fast and accurate emission models are essential for engine model-based control and for engine calibration and optimization using hardware-in-the-loop (HIL) setups. In this study, a fast-response NOx emissions sensor is used to measure the transient NOx emissions from a dual-fuel hydrogen-diesel engine. Subsequently, steady-state models (SSMs), quasi steady-state models (QSSMs), and transient sequential models (TSMs) in the form of black-box (BB) and gray-box (GB) models are developed for transient NOx emissions prediction. GB models utilize both information from a one dimensional (1D) physical engine model and experimental data for training, while BB models only use experimental data. SSMs are optimized artificial neural networks (ANNs) trained using steady-state data, QSSMs are optimized ANNs trained using transient data, and TSMs are time-series networks trained using transient data. Long short-term memory (LSTM) and gated recurrent unit (GRU) networks are used as the time-series deep learning networks. The results showed that the 1D physical model has the poorest performance with successive model performance improvement from SSM to QSSM and from QSSM to TSM. The developed BB TSM model in this study can predict transient NOx emissions with an R2 value greater than 0.96 at 89,000 predictions per second which makes this model suitable for real-time engine model-based control where computational efficiency is crucial. The developed GB TSM model can predict transient NOx emissions with an R2 value greater than 0.97 but it is computationally more expensive. The extra accuracy of the GB TSM models makes them the best choice for HIL setups where more computational power is available, and accuracy is more crucial.