Model Assessment Research Articles

Food Web Bioaccumulation Model for Ecological Risk Assessment of Emerging Organic Pollutants in Marine Ecosystems: Principles, Advances and Challenges

The bioaccumulation and trophic transfer of pollutants in marine ecosystem members determine their ultimate ecological risks. Food web bioaccumulation models are widely used in scientific and regulatory programs to assess the bioaccumulation and ecological risks of pollutants at the ecosystem scale. The food web models are mainly established through concentration- and fugacity-based modeling approaches and include some chemical, food web-related, physiological and environmental factors. The models applied in the “forward approach” predict bioaccumulation and conduct internal exposure level-based ecological risk assessment (IEL-ERA), whereas those in the “reverse approach” are used to back-calculate the IEL-based predicted no-effect concentrations (PNECs) or environmental criteria. However, some challenges still exist in the application of food web model integrated risk assessment, including the lack of standardized/generalized frameworks, the lack of chemical- and species-specific toxicokinetic data and internal exposure (or tissue residue)-based toxicity data, and the lack of uncertainty-control methods in model estimation and parameterization. There are urgent requirements to improve models, integrate methods and update study designs in the assessment and prediction of “system-scale risks” of marine emerging organic pollutants.

Multi-Label Feature Aware XGBoost Model For Student Performance Assessment Using Behavior Data in Online Learning Environment

In light of recent outbreaks like COVID19, the use of online-based learning streams (i.e., e-Learning systems) has increased significantly. Institutional efforts to boost student achievement have made precise predictions of academic success a priority. To analyze student sessions-streams and anticipate academic success, e-Learning platforms are starting to combine Datamining (DM) with Machine-Learning (ML) techniques. Recent research highlights the difficulties that ML-based methods have while dealing with unbalanced data. In tackling ensemble-learning, we combine several ML algorithms to select the most appropriate approach for the given data. Current ensemble-based approaches for predicting student achievement, nevertheless, don't do exceptionally well, particularly when it comes to multi-label classification, because they don't factor the relevance of features into their approaches. This study presents multi-label feature aware XGBoost (MLFA-XGB) method that improves upon the previously used ensemble-learning technique. The MLFA-XGB makes use of a robust cross-validation approach for gaining a deeper understanding of feature relationships. The experimental results demonstrate that in comparison with the state-of-the-art ensemble-based student achievement predictive approach, this suggested MLFA-XGB based approach provides much higher accuracy for prediction.

Assessment of joint distributions of wave heights and periods

Short-term joint distributions of wave heights and periods provide information about the sea surface that is needed for engineering applications. In recent years much effort has been invested in predicting and validating the (univariate) distributions of wave heights; however, the same cannot be said about the joint probability of wave heights and periods. Comparisons of the joint distribution models with field observations are quite limited in terms of both the duration of the data and the number of locations considered. In addition, the often-contradicting demands for longer sea-state records and stationarity in the datasets have presented researchers with a bit of a conundrum. It is likely that some previous studies that aggregated data from several non-stationary sea states might have yielded improper inferences. The current study attempts a robust assessment of the available theoretical distribution models utilizing an extensive dataset of wave measurements around the UK coast. The results indicate that the Zhang and Guedes Soares (2106) model generally performs the best, and is perhaps a notable improvement over the somewhat bleak outlook laid out for these types of models in earlier years. The results also suggest that the discrepancy between the models and the data arises primarily from the poor ability of the marginal period distributions to capture reality.

Comparative life cycle assessment and route optimization modeling of smart versus conventional municipal waste collection: Environmental impact analysis in an urban context

Comparative life cycle assessment and route optimization modeling of smart versus conventional municipal waste collection: Environmental impact analysis in an urban context

Hydrological drought assessment of the Yellow River Basin based on non-stationary model

Study Regionthe Yellow River Basin Study FocusNon-stationary hydrological conditions are becoming increasingly common due to climate change and human activities, and pose a novel challenge to the management of water resources and related risks, especially for large basins. However, prevalent research on drought assessment often ignores the non-stationary characteristics of hydrological processes. In this study, we investigated the stationarity of the runoff of the Yellow River Basin (YRB), the second-longest basin in China. We used an approach for assessing non-stationary droughts based on the generalized additive model for location, scale, and shape to establish a standardized runoff index containing covariates (SRI_cov) to identify hydrological droughts in the basin from 1986 to 2015. New Hydrological Insights for the RegionThe results show that the runoff was non-stationary in the YRB. Based on SRI_cov, hydrological drought predominantly occurred in the entire YRB in spring. From 1986, the number of months in which droughts occurred in the YRB exhibited a general trend of increase and peaked around 2002. After that, the total number of droughts significantly decreased but extreme droughts had become more prominent since 2005. The drought was more severe in the middle reaches of the Yellow River, and was characterized by a high frequency, intensity, and severity. Our analysis enhances the understanding of hydrological modeling and drought assessment under non-stationary conditions.

High-resolution 3D hydrodynamic modeling and comprehensive assessment of tidal current energy resources

As a clean and renewable resource, tidal current holds significant development potential. However, accurately assessing tidal current energy resources has been a persistent challenge. This study proposes a comprehensive assessment method using a small-scale, high-precision, multi-criteria three-dimensional hydrodynamic model to assess resource reserves applied to the Zhairuoshan Waterway. The hydrodynamic model was employed to simulate critical parameters such as tidal level, flow velocity, and flow direction. Multiple statistical indicators were used to evaluate the model's accuracy, showing consistency. The study analysed the flow speed distribution over a complete cycle, revealing an average flow speed of 1.51 m/s, with a maximum speed reaching 3.83 m/s. The generally high flow speeds within the waterway, coupled with the stability of the tidal current, indicate the significant resource potential of the study area. A quantitative analysis of the spatial distribution of tidal current energy was conducted using energy flux density, and the results showed that high energy density areas are widely distributed. Furthermore, the theoretical tidal energy resources of the waterway were calculated, showing that the maximum resource reserves could reach 9.98 MW at the selected cross-sections. Coupled with the analysis of unit-width energy flux, the high-potential areas within the waterway were further identified.

Dynamic model and performance assessment of inter-story isolation system with supplement inerter (IISI)

The hybrid control combining the conventional seismic isolation and tuned mass damper inerter is recently gaining popularity for the seismic protective strategy. This paper presents a simplified dynamic model of the inter-story isolation system with a supplement inerter (IISI) for seismic performance assessment. The dynamic characteristics, such as modal frequencies, mode shapes, damping ratios, and participation factors, as well as the seismic response spectrum, were studied considering the importance of the equivalent linear model for seismic design. The vibration characteristic formula of the IISI was derived using a simplified 2-degree-of-freedom (2DOF) model. The influence of the calculated parameters of the model on the vibration characteristics was analyzed. To understand the isolation control mechanism, the response transfer function in the frequency domain was derived, and the effects of model parameters on the deformation amplitude amplification factor of the superstructure and substructure were analyzed. Based on equivalent linear properties, a seismic response prediction formula based on the response spectrum method is provided and verified by time history analysis (THA). The effect of soil-structure interaction (SSI) on the simplified 4-DOF model was also studied. The seismic responses and effectiveness of the IISI were evaluated using a Multi-degree-of-freedom (MDOF) model example. The research results demonstrate that the addition of an inerter changes the vibration characteristics of the traditional isolation system. The supplement inerter is effective in controlling the seismic response of the IISI, and the established analysis method can be used to analyze the seismic response of the IISI by incorporating additional inerter mounted in the isolation layer. The hybrid control strategy shows better effectiveness in reducing displacement in the isolation layer.

Hydrogeochemical properties, source provenance, distribution, and health risk of high fluoride groundwater: Geochemical control, and source apportionment

This study evaluated high fluoride (F-) levels, source distribution, provenance, health risk, and source apportionment in the groundwater of Sargodha, Pakistan. Therefore, 48 groundwater samples were collected and analyzed by ion-chromatography (DX-120, Dionex). The lowest concentration of F- was 0.1, and the highest was 5.8 mg/L in the aquifers. In this study, 43.76% of the samples had exceeded the World Health Organization's allowable limit of 1.5 mg/L. The hydrogeochemical facies in Na-rich and Ca-poor aquifers showed NaCl (66.6%), NaHCO3 (14.5%), mixed CaNaHCO3 (8.3%), CaCl2 (8.3%), mixed CaMgCl2 (2%), and CaHCO3 (2%) type water. Alkaline pH, high Na+, HCO3- concentrations, and poor Ca-aquifers promoted F- dissolution in aquifer. The significant positive correlations between Na⁺ and F- suggested cation exchange, where elevated Na⁺ occurs in Ca-poor aquifers. The cation exchange reduces the availability of Ca2+ would lead to higher F- concentrations. Meanwhile, the correlation between HCO₃- and F- indicates that carbonate minerals dissolution helps in increasing pH and HCO₃- as a result F- triggers in aquifers. Groundwater chemistry is primarily governed by the weathering of rock, water-rock interaction, ion-exchange, and mineral dissolution significantly control groundwater compositions. Cluster analysis (CA) determined three potential clusters: less polluted (10.4%), moderately polluted (39.5%), and severely polluted (50%) revealing fluoride toxicity and vulnerability in groundwater wells. Mineral phases showed undersaturation and saturation determining dissolution of minerals and precipitation of minerals in the aquifer. PCAMLR model determined that high fluoride groundwater takes its genesis from F-bearing minerals, ion exchange, rock-water interaction, and industrial, and agricultural practices. The health risk assessment model revealed that children are at higher risk to F- toxicity than adults. Thus, groundwater of the area is unsuitable for drinking, domestic, and agricultural needs.

Network-based resilience assessment of an urban rail transit infrastructure with a multi-dimensional performance metric

With urban and public transportation development, the urban rail transit networks (URTN) become increasingly complex, evolving into multi-layer and intricate systems. The operating environment is complex, and the rising frequency of emergencies significantly impacts the entire network. Consequently, the resilient ability of URTN to deal with risk attacks has become an important research field. In this paper, a bilayer URTN model is constructed, encompassing urban and suburban rail transit, proposing the adjacency matrix and the coupling mechanism of subnetworks. Then, the performance of a URTN jointly considering network efficiency, network structure entropy, and the number of trips per unit time is addressed. By considering passenger travel alternatives under disruptions, we develop an improved Logit stochastic user equilibrium (SUE) passenger assignment model based on generalized travel time costs, introducing the failure degree of the stations, to predict passenger flow paths. A multi-dimensional resilience assessment model is proposed using the resilience change curve for attack scenarios and recovery strategies, and a comprehensive network performance system through the AHP method to determine the weight coefficients of indicators. Finally, we present a case study based on the Shanghai URT network to explore the resilience of URTN in the morning peak and simulate the network resilience under different failure scenarios. The results indicate that the proposed model can assess the resilience of the Shanghai MRTN under random and malicious failure, and the former has less comprehensive performance losses and greater resilience. Additionally, the sensitivity analysis is conducted to explore the impact of the number of failed stations, the failure degree of stations, and recovery ability on network resilience. This study provides theoretical support for urban managers and builders regarding network structure and emergency management, and offers relevant suggestions.

Life cycle assessment of the climate change impact of magnesium phosphate cements formulated with tundish deskulling waste compared to conventional cement

Ordinary Portland cement (OPC) production significantly contributes to greenhouse gas emissions due to high resource consumption and CO2 output. It is therefore imperative to investigate alternative cements, such as magnesium phosphate cement (MPC), as a potential solution. This study is based on Life Cycle Assessment (LCA) methodology, comparing OPC with alternative magnesium phosphate cements (MPC) developed at the laboratory scale. The novelty of this study considers two types of alternative cements that use two different sources of MgO: MPC-MgO, developed with pure MgO, and MPC-TUN, formulated using tundish deskulling waste from steelmaking industry.The evaluated functional units are 1 tonne of cement, 1 m3 of cement paste, and 1 m3 of mortar, all of them are designed for the same function, which is as non-structural precast elements. The study assesses climate change impacts under two future scenarios: 1) electricity decarbonisation in the background economy using projections from Integrated Assessment Models and 2) electricity decarbonisation and a fuel switch in the cement kilns.The results indicate that MPC-TUN exhibits a lower impact of climate change in terms of CO2 emissions across all functional units and scenarios compared to the other materials. In the most ambitious climate scenario, MPC-TUN mortar exhibits 42% and 56% lower climate change impacts than OPC-CEM I and MPC-MgO mortars, respectively, demonstrating its potential as a more sustainable construction material. Although further research is needed on the applicability of MPC-TUN in construction, regulatory frameworks are advised to simplify barriers to expedite the adoption of sustainable alternative cements.

European anchovy's abundance, more affected by climatic conditions than fishing activities in the northwest African waters.

The marine waters off the coast of northwest Africa are known for being highly productive upwelling regions in the Eastern Atlantic Ocean. The present study evaluated the combined effects of climate change and sustainable fishing levels on the long-term sustainability of European anchovy (Engraulis encrasicolus), a target pelagic species found along the West African coastal waters. The present study used survey biomass time series from survey vessels and species catch time series from commercial fisheries operating in the region. Hence, we investigated the relationship between fishing dynamics and environmental factors. The status of anchovy stock was determined by analyzing changes in different initial depletion rates using a state-space Bayesian surplus production model. We used Principal component analysis to identify the essential environmental data set, while Generalized Additive Models (GAM) were employed to determine the influence of the environment on fisheries. The stock assessment model revealed that fishing intensity did not affect the abundance of anchovy in the region even when tested under different biomass depletion scenarios. However, by employing GAMs, the research assessed the impact of environmental variables on fish biomass, indicating that sea surface temperature (SST) significantly influences anchovy abundance in the region. The latitudinal gradient also had a significant influence on this species' abundance. These findings underscore the vulnerability of the small-scale fisheries sector in the tropical upwelling zone to climate change. The study concludes that even in the face of potential negative impacts from environmental factors, optimizing fishing methods can contribute to the preservation of pelagic resources in the region. It emphasizes the importance of managing fishery resources in a sustainable, economically viable, and socially acceptable way.

Stochastic assessment of voltage sags and techno-economic optimization

Power quality has become an essential concern for industries and domestic energy consumption with the development of sophisticated equipment like micro-electronic devices. When a power quality event happens, such as voltage sag, financial losses for industries increase tremendously. Therefore, voltage sag evaluation becomes crucial for predicting the number of events and severity of voltage sag events of the electrical network. Industries can determine the mitigation investment needed for optimal operation using the knowledge obtained through voltage sag analysis method. Thus, the first part of the study addressed the development of a stochastic assessment method for voltage sag at an interested bus in the power system network, which was an initial objective of the study. The second part of the research investigates dynamic voltage restorer (DVR) control techniques, as well as the necessary capacitance and energy storage dimensions. The studies also recommend a technique for economic evaluation that emphasizes the expense of voltage sag events as well as the cost of investment in DVR with and without voltage sag events. The size of DVR and economic assessment model established in these studies enables industries to rate the cost of voltage sag and the total investment in mitigation devices.

High-traversability and efficient path optimization for deep-sea mining vehicles considering complex seabed environmental factors

In cobalt-rich crust regions, rapid and safe pathfinding is critical for the operation of Deep-Sea Mining Vehicles (DSMVs). However, existing pathfinding solutions rarely account for traversability predictions in complex mining environments and extensive map data often results in computational inefficiencies. This paper proposes a High-Traversability and Efficient Path Optimization (HTEPO) strategy designed to ensure safe navigation and rapid decision-making accounting for complex seabed environmental factors. Initially, we develop a traversability assessment model combining the Analytic Hierarchy Process (AHP) with the Fuzzy Comprehensive Evaluation (FCE) method. This model assigns traversability scores to map out high-risk and impassable areas effectively. We then introduce a novel map model that integrates these scores with Voronoi diagrams to reduce search space and improve computing efficiency, ensuring safer route selections. Further, based on traversability index and map model, an enhanced A* algorithm equipped with adaptive heuristic weights is utilized to swiftly identify high-traversability paths for the DSMVs. Finally, three case studies with different seabed mining environments validate the excellent performance of the proposed method over other major existing methods in terms of path safety and computational efficiency.

Preventative and curative treatment of venous thromboembolic disease in cancer patients

Cancer-associated venous thromboembolism (CAT) is common in patients with cancer and associated with significant morbidity and mortality. The incidence of CAT continues to rise, complicating patient care and burdening healthcare systems. Patients with cancer experiencing VTE face poorer prognoses, making prevention and effective management imperative. This narrative review synthesizes evidence on thromboprophylaxis in ambulatory patients with cancer receiving systemic therapy and acute treatment strategies for CAT. Risk assessment models (e.g., Khorana score) aid in identifying high-risk patients who may benefit from thromboprophylaxis. Pharmacological thromboprophylaxis with low molecular weight heparins (LMWHs) direct oral anticoagulants (DOACs) has been shown to reduce the risk of CAT without significantly increasing the risk of bleeding complications. However, implementation of risk-based strategies remains limited in clinical practice. For acute CAT management, LMWHs have been the standard of care, but DOACs are increasingly favored due to their convenience and efficacy. However, challenges persist, including bleeding risks and drug interactions. Emerging therapies targeting Factor XI inhibitors present promising alternatives, potentially addressing current limitations in anticoagulation management for CAT.

Simulating future cultivated land using a localized SSPs-RCPs framework: A case study in Yangtze River Economic Belt

Simulating cultivated land changes is vital for improving decision-making in sustainable agriculture. Compared to previous studies, the Shared Socioeconomic Pathways and Representative Concentration Pathways (SSPs-RCPs) scenarios provide a comprehensive framework for integrating climate and socioeconomic factors, offering deeper insights into future cultivated land dynamics. Taking the Yangtze River Economic Belt (YREB) as the study area, we developed localized SSPs-RCPs scenarios and adopted Global Change Assessment Model (GCAM) and Future Land Use Simulation (FLUS) model to simulate cultivated land changes under multiple scenarios. The findings indicate that cultivated land is predicted to decrease under all scenarios, with the SSP1-2.6 and SSP3-7.0 scenarios showing the highest and lowest decrease of 39.24% and 8.32%, respectively. By 2050, the loss of cultivated land in the upper reaches is expected to be significantly greater than that in the middle and lower reaches. Furthermore, grasslands, forests, and urban areas are the main types of land that replacing cultivated land in the YREB. The hotspots of cultivated land loss are in mountainous areas such as Sichuan, Yunnan, and Guizhou. Additionally, the gravity center of cultivated land distribution moved along the southeast to northeast, with the center of gravity shifting from Guizhou to Hunan under scenarios SSP1-2.6 and SSP5-8.5. This study offers insights into the future development of cultivated land in the YREB and highlights the significance of implementing and improving policies for its protection.

A food-web assessment model for marine mammals, fish, and fisheries in the Norwegian and Barents Seas

The Norwegian and Barents Seas host large commercial fish populations that interact with each other, as well as marine mammal populations that feed on plankton and fish. Quantifying the past dynamics of these interacting species, and of the associated fisheries in the Norwegian and Barents Sea is of high relevance to support ecosystem-based management. The purpose of this work is to develop a food-web model of intermediate complexity and perform a quantitative assessment of the Norwegian and Barents Sea ecosystems in the period 1988–2021 in a manner that is consistent with existing data and expert knowledge, and that is internally coherent. For this purpose, we use the modelling framework of chance and necessity (CaN). The model construction follows an iterative process that allows to confront, discuss, and resolve multiple issues as well as to recognise uncertainties in expert knowledge, data, and input parameters. We show that it is possible to reconstruct the past dynamics of the food-web only if recognising that some data and assumptions are more uncertain than originally thought. According to this assessment, consumption by commercial fish and catch by fisheries jointly increased until the early 2010 s, after which consumption by fish declined and catches by fisheries stabilised. On an annual basis, fish have consumed an average of 135.5 million tonnes of resources (including 9.5 million tonnes of fish), marine mammals have consumed an average of 22 million tonnes of which 50 % (11 million tonnes) were fish, and fisheries and hunting have captured an average of 4.4 million tonnes and 7 thousand tonnes of fish and marine mammals respectively.

Data-driven model assessment: A comparative study for ship response determination

Several machine learning approaches to determine ship responses via data-driven models have been applied. Input features and parameters used relied on time-series analyses obtained from computational-fluid-dynamics approach. As inputs into the data-driven model, the heave and pitch motions of a ship advancing in irregular, i.e., natural seaway were considered. By using different models in the framework of machine learning, required computation for the associated ship motions may be avoided, thus reducing the computational effort to forecast ship motions. Comparative predictions with numerical simulations revealed that the deep-neural-network method for training in auto-machine-learning instructions yielded the highest accuracy in heave motion, resulting in a non-normalized mean-absolute-error of 0.74, against the corresponding error of 1.07 from numerical computations, whereas the method trained with the tree-based models (the extreme gradient boosting and the Hist gradient boosting regressor) predicted less accurate motions for the tested ship. The model trained with the random forest regressor exhibited an error of 1.10. Numerical simulation based on a field method proved to be the most suitable choice for pitch motion. Despite the few samples available to train the regressors, results demonstrated that the measured data was sufficient to assess the developed data-driven model for ship response determination.

Hyperplasia of fat-containing cells with mature adipocyte marker is associated with pancreatic fat enlargement

Abstract Objectives: The purpose was to elucidate the specific characteristics of fat-containing cells in the pancreas and the mechanism of intra-pancreatic fat deposition (IPFD) in humans. Methods: 15 Japanese patients who had undergone pancreatic resection were enrolled and the normal region of pancreatic samples from each patient was examined. To evaluate the characteristics of fat-containing cells, immunostaining for adiponectin and perilipin 1 was performed and the relationships between the pancreatic fat-cell area or clinical parameters, and the density or the diameter of the fat cells was analyzed. Results: Expression of adiponectin in the cytoplasm and perilipin 1 along the plasma membrane were observed in fat-containing cells in the pancreas. The fat-containing cell area had a significant positive correlation with cell density (r = 0.58, p = 0.023). In addition, fat-containing cell density was significantly positively correlated with homeostasis model assessment insulin resistance (HOMA-IR) (r = 0.61, p = 0.045). Fat-containing cell area was not significantly correlated with cell diameter (r = 0.45, p = 0.095). The diameter of fat-containing cells had significant positive correlations with BMI (r = 0.79, p < 0.001), fasting immunoreactive insulin (r = 0.65, p = 0.029) and HOMA-IR (r = 0.80, p = 0.003). Of all fat-containing cells, 10.4% were intralobular cells, and the diameter of intralobular cells showed a tendency for positive correlation with age (r = 0.52, p = 0.057). Conclusions: The characteristics of fat-containing cells in the pancreas indicate that some of them may be mature adipocytes, and fat volume may be increased by hyperplasia of fat-containing cells associated with insulin resistance. Key terms: pancreatic fat, adipocytes markers, hyperplasia of fat-containing cells, insulin resistance.

Empirical and Machine Learning Models for Soil Erosion Risk Assessment: A Case Study of Tsageri Municipality, Georgia

Soil erosion caused by water is one of the most common causes of land degradation worldwide. Within framework of this research soil erosion risk in Tsageri municipality, Georgia was evaluated using Revised Universal Soil Loss Equation (RUSLE) and a machine learning-based Random Forest (RF) model. Open access digital datasets and field observations collected in 2023-2024, which included visually identified erosion areas and GPS-recorded data on the presence or absence of erosion, were utilized in modeling process. Data processing and modeling conducted using ArcGIS Pro 3.0 and RStudio software. According to RUSLE results, 39.7% of the study area falls under the very low erosion risk zone, and 20.7% is in the very high risk zone. The RF model results indicated that 16.5% of the territory is under very low risk of erosion and 13.9% - very high risk. It was observed that RUSLE model tends to overestimate erosion rates on steep, forested slopes, while the RF model, by incorporating additional variables, provided more accurate prediction. These findings suggest that combining RUSLE with machine learning improves soil erosion risk assessment, particularly in complex landscapes such as in Tsageri municipality. Future researche should focus on testing additional variables to refine the modeling process further and enhance predictions. The generated digital thematic maps offer valuable insights for understanding the spatial dynamics of soil erosion within the study area, analyzing the factors driving the process and developing effective mitigation strategies.

Anti-VEGF treatment outcome prediction based on optical coherence tomography images in neovascular age-related macular degeneration using a deep neural network.

Age-related macular degeneration (AMD) is a major cause of blindness in developed countries, and the number of affected patients is increasing worldwide. Intravitreal injections of anti-vascular endothelial growth factor (VEGF) are the standard therapy for neovascular AMD (nAMD), and optical coherence tomography (OCT) is a crucial tool for evaluating the anatomical condition of the macula. However, OCT has limitations in accurately predicting the degree of functional and morphological improvement following intravitreal injections. Artificial intelligence (AI) has been proposed as a tool for predicting the treatment response of nAMD based on OCT biomarkers. Our study focuses on the development and assessment of an AI model utilizing the DenseNet201 algorithm. The model aims to predict anatomical improvement based on OCT images before, and during anti-VEGF therapy. The training process involves two scenarios: (1) using only preinjection OCT images and (2) utilizing both OCT images before and during anti-VEGF therapy for model training. The outcomes of our investigation, involving 2068 images from a cohort of 517 Korean patients diagnosed with nAMD, indicate that the AI model we introduced surpassed the predictive performance of ophthalmologists. The model exhibited a sensitivity of 0.915, specificity of 0.426, and accuracy of 0.820. Notably, its predictive capabilities were further enhanced with the inclusion of additional OCT images taken after the first and second injections during the loading phase. The treatment prediction performance of the model was the highest when using all input modalities (before injection, and after the first and second injections) and concatenation-based fusion layers. This study highlights the potential of AI in assisting individualized and tailored nAMD treatment.