Model Overfitting Research Articles

A novel hybrid inception-xception convolutional neural network for efficient plant disease classification and detection

Plants are essential at all stages of living things. Plant pests, diseases, and symptoms are most regularly visible in plant leaves and fruits and sometimes within the roots. Yet, their diagnosis by experts in the laboratory is expensive, tedious, and time-consuming if the samples involve laboratory analysis. Failure to detect early plant symptoms and diseases is the core biotic cause of increased plant stresses, structure, health, reduced subsistence farming, and threats to global food security. To mitigate these problems at a social, economic, and environmental level, inappropriate herbicide application reduction and early plant disease detection and classification (PDDC) are significant solutions in this case. Advancements in transfer learning techniques have resulted in effective results in smart farming and have become extensively used in disease identification and classification research studies. This study presents a novel hybrid inception-xception (IX) using a convolution neural network (CNN). The presented model combines inception and depth-separable convolution layers to capture multiple-scale features while reducing model complexity and overfitting. In contrast to ordinary CNN architectures, it extends the network for better feature extraction, improving PDDC performance that demands diverse feature competencies. It further presents a real-time artificial intelligence (AI) application available in MATLAB, Android, and Servlet to automatically identify and classify diseases based on the leaf environment using improved CNN, machine learning (ML), and computer vision techniques. To assess the presented IX-CNN model performance, different classifiers, namely, support vector machine (SVM), decision tree (DT) and random forest (RF), were used. The experiments used six datasets, including PlantVillage, Turkey Disease, Plant Doc, Rice Disease, RoCole, and NLB datasets. Plant Doc, PlantVillage, and Turkey Disease datasets demonstrated an accuracy of 100%. Rice Disease, RoCole, and NLB attained an accuracy of 99.79%, 99.95%, and 98.64%, respectively.

Leveraging Incremental Machine Learning for Reconfigurable Systems Modeling under Dynamic Workloads

Dynamic workload orchestration is one of the main concerns when working with heterogeneous computing infrastructures in the edge-cloud continuum. In this context, FPGA-based computing nodes can take advantage of their improved flexibility, performance and energy efficiency provided that they use proper resource management strategies. In this regard, many state-of-the-art systems rely on proactive power management techniques and task scheduling decisions, which in turn require deep knowledge about the applications to be accelerated and the actual response of the target reconfigurable fabrics when executing them. While acquiring this knowledge at design time was more or less feasible in the past, with applications mostly being static task graphs that did not change at run time, the highly dynamic nature of current workloads in the edge-cloud continuum, where tasks can be deployed on any node and at any time, has removed this possibility. As a result, being able to derive such information at run time to make informed decisions has become a must. This paper presents an infrastructure to build incremental ML models that can be used to obtain run-time power consumption and performance estimations in FPGA-based reconfigurable multi-accelerator systems operating under dynamic workloads. The proposed infrastructure features a novel stop-and-restart resource-aware mechanism to monitor and control the model training and evaluation stages during normal system operation, enabling low-overhead updates in the models to account for either unexpected acceleration requests (i.e., tasks not considered previously by the models) or model drift (e.g., fabric degradation). Experimental results show that the proposed approach induces a maximal additional error of 3.66% compared to a continuous training alternative. Furthermore, the proposed approach incurs only a 4.49% execution time overhead, compared to the 20.91% overhead induced by the continuous training alternative. The proposed modeling strategy enables innovative scheduling approaches in reconfigurable systems. This is exemplified by the conflict-aware scheduler introduced in this work, which achieves up to a 1.35 times speedup in executing the experimental workload. Additionally, the proposed approach demonstrates superior adaptability compared to other methods in the literature, particularly in response to significant changes in workload and to mitigate the effects of model overfitting. The portability of the proposed modeling methodology and monitoring infrastructure is also shown through their application to both Zynq-7000 and Zynq UltraScale+ devices.

Adversarial Robustness for Deep Learning-Based Wildfire Prediction Models

Rapidly growing wildfires have recently devastated societal assets, exposing a critical need for early warning systems to expedite relief efforts. Smoke detection using camera-based Deep Neural Networks (DNNs) offers a promising solution for wildfire prediction. However, the rarity of smoke across time and space limits training data, raising model overfitting and bias concerns. Current DNNs, primarily Convolutional Neural Networks (CNNs) and transformers, complicate robustness evaluation due to architectural differences. To address these challenges, we introduce WARP (Wildfire Adversarial Robustness Procedure), the first model-agnostic framework for evaluating wildfire detection models’ adversarial robustness. WARP addresses inherent limitations in data diversity by generating adversarial examples through image-global and -local perturbations. Global and local attacks superimpose Gaussian noise and PNG patches onto image inputs, respectively; this suits both CNNs and transformers while generating realistic adversarial scenarios. Using WARP, we assessed real-time CNNs and Transformers, uncovering key vulnerabilities. At times, transformers exhibited over 70% precision degradation under global attacks, while both models generally struggled to differentiate cloud-like PNG patches from real smoke during local attacks. To enhance model robustness, we proposed four wildfire-oriented data augmentation techniques based on WARP’s methodology and results, which diversify smoke image data and improve model precision and robustness. These advancements represent a substantial step toward developing a reliable early wildfire warning system, which may be our first safeguard against wildfire destruction.

Inference of Behavioral Decision-Making Using AI and Statistical Data Analysis in the Big Data Environment

Big data are of essence to enhance the accuracy of making decisions and in improving the ability to forecast. However, working with large and complicated datasets, the traditional methods of analyzing data tend to be grossly inadequate. It is at this point, amongst others, that artificial intelligence has now presented a feasible solution to such limitations, especially with the model called machine learning. Based on that, this research will look into the aspect of integration between artificial intelligence and statistical methods of analysis in inferring behavioral decisions from big data. This work considers a number of datasets related to marketing, health care, and finance, comparing the efficiency of the application of a range of artificial intelligence (AI) models, especially random forests, Long Short Term Memory (LSTM), and convolutional neural network (CNN) algorithms, against the classical statistical ones. Standard performance evaluation indicators apply: accuracy, precision, recall rate, F1 score are applied in the model. Results have shown that, though model interpretability and overfitting are challenges, the predictive accuracy of artificial intelligence models is somewhat better in comparison with conventional statistical methods. The present study underlines the transformative potential of AI in changing decision-making across many industries but also highlights key areas for further improvement on behalf of real-time processing abilities and ethical deployment consideration.

Optimization of the Effect of Network Information Flow Advertising Based on User Behavior Data Encryption and Deep Learning

Existing network information flow advertising mostly relies on user behavior data, which is easy to cause user privacy leakage and low precision of advertising delivery. This article combines user behavior data encryption and deep learning technology to propose a method for optimizing the effect of network information flow advertising. The user behavior data is encrypted using Paillier homomorphic encryption technology. Then, the stacked auto encoder (SAE) is used to extract high-order nonlinear features, and the shallow features are automatically combined based on Field-aware Factorization Machines (FFM) to construct a SAE-FFM comprehensive model. At the same time, dropout is used to reduce model overfitting, and the advertising effect is optimized through two-stage pre-training and fine-tuning. The experimental results show that when the number of impressions is 10,000, the average advertising click through rate and conversion rate of the SAE-FFM model are 4.50% and 0.59% respectively, and the information leakage rate is 0.052%. The results show that the SAE-FFM model can greatly improve the click through rate and conversion rate of advertisements, prevent privacy leakage, and help the development of the information flow advertising industry.

An Analysis of the Application and Impact Mechanisms of Machine Learning in the Field of Asset Pricing

Issues related to asset pricing and return forecasting have consistently been prominent research topics in both the financial industry and academia. With the rapid advancement of science, technology, and artificial intelligence, machine learning has garnered extensive attention from scholars for its robust self-learning and adaptive capabilities, as well as its exceptional advantages in large-scale data processing. This paper provides a detailed overview of the application of emerging machine learning methods in asset pricing and offers an in-depth analysis of machine learnings contributions to this field based on existing research. This research finds that machine learning has introduced profound changes and progressive innovations in asset pricing models. Additionally, this paper examines the limitations of machine learning methods, objectively highlighting current shortcomings such as poor interpretability and model overfitting. Finally, this paper proposes future directions for improving machine learning models to advance the financial field further.

Deep Learning for Multi-Label Facial Attribute Classification on Large-Scale Image Datasets (CelebA)

The exponential growth of data in recent years has led to an increasing demand for advanced techniques, especially those that work on large and complex data. This has given deep learning a significant advance in dealing with the tasks of analyzing, improving, and distinguishing big data. Our research focused on CNNs from this data and applying deep learning algorithms and their analysis to a large-scale image dataset. More specifically, our research focused on a dataset called CelebA, which contains more than 200,000 face images annotated with 40 binary facial features. It is a multi-label classification model based on the ResNet-50 architecture that has been fine-tuned to predict different facial features and hair color such as age, gender, and facial expressions. It was also trained using data augmentation, taking into account pose differences and background clutter to reduce imbalance between classes. These results reflect very strong predictive performance, with an average mean accuracy of 0.86 and an overall F1 score of 0.81 across all features. Attributes identified by clear visual cues—for example, “smiling,” “male ”and“ wearing lipstick”—were highly accurate, while less obvious attributes such as “big lips” and “narrow eyes” were more difficult to classify. We would like to point out that the results demonstrate the high efficiency of using deep learning models for multi-label classification on big data while solving problems associated with class imbalance and overfitting models. This research leads to the larger general field of big data analytics; in particular, it demonstrates how deep learning can be efficiently applied to large image datasets for automatic attribute recognition. It also opens up potential applications in areas such as biometric identification, surveillance, and human-computer interaction.

Comparison of Different Machine Learning Methodologies for Predicting the Non-Specific Treatment Response in Placebo Controlled Major Depressive Disorder Clinical Trials.

Placebo effect represents a serious confounder for the assessment of treatment effect to the extent that it has become increasingly difficult to develop antidepressant medications appropriate for outperforming placebo. Treatment effect in randomized, placebo-controlled trials, is usually estimated by the mean baseline adjusted difference of treatment response in active and placebo arms and is function of treatment-specific and non-specific effects. The non-specific treatment effect varies subject by subject conditional to the individual propensity to respond to placebo. This effect is not estimable at an individual level using the conventional parallel-group study design, since each subject enrolled in the trial is assigned to receive either active treatment or placebo, but not both. The objective of this study was to conduct a comparative analysis of the machine learning methodologies to estimate the individual probability of a non-specific treatment effect. The estimated probability is expected to support novel methodological approaches for better controlling effect of excessively high placebo response. At this purpose, six machine learning methodologies (gradient boosting machine, lasso regression, logistic regression, support vector machines, k-nearest neighbors, and random forests) were compared to the multilayer perceptrons artificial neural network (ANN) methodology for predicting the probability of individual non-specific treatment response. ANN achieved the highest overall accuracy among all methods tested. A fivefold cross-validation was used to assess performances and risks of overfitting of the ANN model. The analysis conducted without subjects with non-specific effect indicated a significant increase of signal detection with significant increase in effect size.

DIFLF: A domain-invariant features learning framework for single-source domain generalization in mammogram classification.

Single-source domain generalization (SSDG) aims to generalize a deep learning (DL) model trained on one source dataset to multiple unseen datasets. This is important for the clinical applications of DL-based models to breast cancer screening, wherein a DL-based model is commonly developed in an institute and then tested in other institutes. One challenge of SSDG is to alleviate the domain shifts using only one domain dataset. The present study proposed a domain-invariant features learning framework (DIFLF) for single-source domain. Specifically, a style-augmentation module (SAM) and a content-style disentanglement module (CSDM) are proposed in DIFLF. SAM includes two different color jitter transforms, which transforms each mammogram in the source domain into two synthesized mammograms with new styles. Thus, it can greatly increase the feature diversity of the source domain, reducing the overfitting of the trained model. CSDM includes three feature disentanglement units, which extracts domain-invariant content (DIC) features by disentangling them from domain-specific style (DSS) features, reducing the influence of the domain shifts resulting from different feature distributions. Our code is available for open access on Github (https://github.com/85675/DIFLF). DIFLF is trained in a private dataset (PRI1), and tested first in another private dataset (PRI2) with similar feature distribution to PRI1 and then tested in two public datasets (INbreast and MIAS) with greatly different feature distributions from PRI1. As revealed by the experiment results, DIFLF presents excellent performance for classifying mammograms in the unseen target datasets of PRI2, INbreast, and MIAS. The accuracy and AUC of DIFLF are 0.917 and 0.928 in PRI2, 0.882 and 0.893 in INbreast, 0.767 and 0.710 in MIAS, respectively. DIFLF can alleviate the influence of domain shifts only using one source dataset. Moreover, DIFLF can achieve an excellent mammogram classification performance even in the unseen datasets with great feature distribution differences from the training dataset.

Prob-sparse self-attention extraction of time-aligned dynamic functional connectivity for ASD diagnosis.

Dynamic functional connectivity (DFC) has shown promise in the diagnosis of Autism Spectrum Disorder (ASD). However, extracting highly discriminative information from the complex DFC matrix remains a challenging task. In this paper, we propose an ASD classification framework PSA-FCN which is based on time-aligned DFC and Prob-Sparse Self-Attention to address this problem. Specifically, we introduce Prob-Sparse Self-Attention to selectively extract global features, and use self-attention distillation as a transition at each layer to capture local patterns and reduce dimensionality. Additionally, we construct a time-aligned DFC matrix to mitigate the time sensitivity of DFC and extend the dataset, thereby alleviating model overfitting. Our model is evaluated on fMRI data from the ABIDE NYU site, and the experimental results demonstrate that the model outperforms other methods in the paper with a classification accuracy of 81.8%. Additionally, our research findings reveal significant variability in the DFC connections of brain regions of ASD patients, including Cuneus (CUN), Lingual gyrus (LING), Superior occipital gyrus (SOG), Posterior cingulate gyrus (PCG), and Precuneus (PCUN), which is consistent with prior research. In summary, our proposed PSA framework shows potential in ASD diagnosis as well as automatic discovery of critical ASD-related biomarkers.

Intelligent Construction Risk Management Through Transfer Learning: Trends, Challenges, and Future Strategies

Construction risk management has evolved significantly by integrating artificial intelligence (AI) technologies, particularly machine learning (ML), to enhance predictive capabilities. Transfer learning (TL), a promising subfield of ML, has the potential to further revolutionize construction safety by enabling models trained in one domain to be adapted for related tasks in construction risk scenarios. This systematic review explores the current trends in applying TL to construction risk management, identifies key challenges, and highlights future opportunities for advancement. The review first assesses TL's ability to mitigate common issues such as data scarcity, overfitting, and lengthy model training times, which often hinder traditional ML approaches. Key challenges include the complexity of domain adaptation, the lack of standardized datasets, and the need for robust validation methods. Despite these barriers, the potential for TL to improve predictive accuracy, efficiency, and cross-project learning makes it a transformative tool. Finally, future research directions are proposed to optimize TL techniques for real-time, intelligent construction risk management systems.

Optimizing malware detection with redundant sample filtering for efficient retraining

ABSTRACT Optimizing the efficiency of ML classifiers for Android malware detection is essential due to the continuous influx of new apps, many of which are redundant. These redundant apps increase time and space complexity, leading to model overfitting and degrading performance during classifier training and retraining. Existing dataset filtering mechanisms often result in slight performance degradation, whereas current retraining methods are unsustainable for longer periods due to their inability to handle growing data volumes. To address these challenges, we propose a novel opcode sequence-based redundant sample filtering mechanism that identifies and eliminates similar apps before retraining. This approach reduced the dataset size by 56%, resulting in a 37% reduction in false predictions for apps belonging to 2018. The classifier required updates only for samples whose opcode sequences significantly differed from those in the training dataset, based on an Ochiai coefficient threshold of 0.4. Consequently, only 8% of apps from 2018 to 2020 were necessary for classifier retraining, improving detection accuracy to 0.94 for apps belonging to 2021. Additionally, the mechanism has proven sustainable, achieving 0.95 detection accuracy for apps belonging to 2024 with an FPR of just 0.02. Furthermore, it identifies 91% of obfuscated malware apps, showcasing robustness against advanced evasion techniques.

Changes in Ginkgo biloba L.’s Habitat Due to Climate Change in China

Ginkgo biloba L. was named by Carl Linnaeus in 1771; a “living fossil” with immense medicinal and conservation value, it is a nationally first-class protected wild plant. However, many Ginkgo populations are under threat from habitat destruction, human exploitation, and over-harvesting, which have limited their numbers and range. Using an optimized MaxEnt model in R, this study analyzed Ginkgo distribution points and 22 ecological factors in China to explore the key environmental factors affecting its geographical distribution. The study also predicted the spatial distribution patterns and centroid changes of potential suitable areas under three different carbon emission pathways: current conditions, 2021–2040 (2030s), 2041–2060 (2050s), and 2061–2080 (2070s). The findings are as follows: (1) The optimal combination of model parameters (RM = 3.2, FC = LPH) reduced model complexity and overfitting and achieved very high prediction accuracy with an optimized AUC value of 0.928. (2) The key environmental factors influencing Ginkgo growth include precipitation in the driest month (20–175 mm), minimum temperature in the coldest month (−4 to 3 °C), precipitation in the hottest quarter (450–2500 mm), and a temperature seasonal variation deviation greater than 580. (3) Under the three future climate scenarios (SSP126, SSP245, and SSP585), the potential suitable habitat area for Ginkgo in China was increased, with the distribution range migrating to higher latitudes, Under the three different development models, the total suitable area followed this order: SSP126 > SSP245 > SSP585. Highly and moderately suitable areas are concentrated in the Yangtze River Basin. This study is highly significant for the ecological protection of Ginkgo, aiding in the rational planning of potential suitable areas, enhancing the monitoring of key conservation areas, and developing effective protection strategies in a timely manner.

Modeling impacts of traffic, air pollution, and weather conditions on cardiopulmonary disease mortality.

Cardiopulmonary disease (CPD) is a leading cause of death worldwide. Increasing evidence shows that air pollution and exposure to weather conditions have important contributory roles. Understanding the interaction of these factors is difficult due to the complexity of the relationship between CPD, air pollution, and environmental factors. This paper uses regression models and machine learning approaches to explore these relationships, and investigate whether meteorological factors and air pollution have a synergistic effect on CPD. We use daily data from 2009-2018 from four cities representing the heterogenous climate conditions in Norway: the far north, the west coast, mid-Norway, and the south-east. We demonstrate the importance of the interaction between weather and air pollution associated with higher CPD mortality, as is exposure to air pollution in the form of particulate matter. This impact is seasonal. Traffic is also positively related to CPD mortality, which may be caused indirectly through increased pollution. We demonstrate that machine learning outperforms regression models in terms of the accuracy of predicting CPD mortality. The inclusion of rich lagged structures and interactions between environmental factors are both important but can lead to overfitting of traditional models; since these cities are not large cities by international standards, it is surprising that environmental factors have such obvious impacts on CPD mortality. CPD mortality shows a clear negative trend, implying an improvement in the public health situation.

Small-Sample Target Detection Across Domains Based on Supervision and Distillation

To address the issues of significant object discrepancies, low similarity, and image noise interference between source and target domains in object detection, we propose a supervised learning approach combined with knowledge distillation. Initially, student and teacher models are jointly trained through supervised and distillation-based approaches, iteratively refining the inter-model weights to mitigate the issue of model overfitting. Secondly, a combined convolutional module is integrated into the feature extraction network of the student model, to minimize redundant computational effort; an explicit visual center module is embedded within the feature pyramid network, to bolster feature representation; and a spatial grouping enhancement module is incorporated into the region proposal network, to mitigate the adverse effects of noise on the outcomes. Ultimately, the model undergoes a comprehensive optimization process that leverages the loss functions originating from both the supervised and knowledge distillation phases. The experimental results demonstrate that this strategy significantly boosts classification and identification accuracy on cross-domain datasets; when compared to the TFA (Task-agnostic Fine-tuning and Adapter), CD-FSOD (Cross-Domain Few-Shot Object Detection) and DeFRCN (Decoupled Faster R-CNN for Few-Shot Object Detection), with sample orders of magnitude 1 and 5, increased the detection accuracy by 1.67% and 1.87%, respectively.

Be aware of overfitting by hyperparameter optimization!

Hyperparameter optimization is very frequently employed in machine learning. However, an optimization of a large space of parameters could result in overfitting of models. In recent studies on solubility prediction the authors collected seven thermodynamic and kinetic solubility datasets from different data sources. They used state-of-the-art graph-based methods and compared models developed for each dataset using different data cleaning protocols and hyperparameter optimization. In our study we showed that hyperparameter optimization did not always result in better models, possibly due to overfitting when using the same statistical measures. Similar results could be calculated using pre-set hyperparameters, reducing the computational effort by around 10,000 times. We also extended the previous analysis by adding a representation learning method based on Natural Language Processing of smiles called Transformer CNN. We show that across all analyzed sets using exactly the same protocol, Transformer CNN provided better results than graph-based methods for 26 out of 28 pairwise comparisons by using only a tiny fraction of time as compared to other methods. Last but not least we stressed the importance of comparing calculation results using exactly the same statistical measures.Scientific Contribution We showed that models with pre-optimized hyperparameters can suffer from overfitting and that using pre-set hyperparameters yields similar performances but four orders faster. Transformer CNN provided significantly higher accuracy compared to other investigated methods.

A multivariable prediction model to identify anti-CCP positive people in those with non-specific musculoskeletal symptoms in primary care.

We aimed to develop a prediction model identifying people presenting to primary care with musculoskeletal symptoms likely to be anti-CCP positive and therefore at risk of developing rheumatoid arthritis (RA). Participants aged ≥16 years, with new-onset non-specific musculoskeletal symptoms and no history of clinical synovitis, completed a symptom questionnaire and had an anti-CCP test. Model development used LASSO-penalised logistic regression, performance was assessed using area under the receiver operating characteristic curve (AUROC) and decision curve analysis, model over-fit was estimated using bootstrapping and cross-validation. Participants were followed-up at 12 months for RA or seronegative/undifferentiated inflammatory arthritis (IA) diagnosis. Analysis included 6879 participants; 203 (2.95%) anti-CCP positive. Eleven predictors were retained: male sex, first-degree relative with RA, ever smoked, and joint pain in: back, neck, shoulders, wrists, hands/fingers, thumbs, knees, feet/toes. AUROC was 0.65 (95% CI:(0.61, 0.69), optimism = 0.03). Using a 4% decision threshold, the model recommended an anti-CCP test in 1288 (18.7%) participants, 78 (6.1%) of whom were anti-CCP positive, compared with 125/5591 (2.2%) below the threshold. Net benefit was 0.0040 (0.0020 corrected). Forty-eight participants were diagnosed with IA/RA within 12 months. Of those who were above the threshold and anti-CCP positive, 32.1% developed IA/RA compared with 0.4% of those who were anti-CCP negative. Of those below the threshold, 0.3% were diagnosed with IA/RA. Targeted anti-CCP testing in primary care may aid earlier identification of people at risk of RA, prompting specialist referral to rheumatology for earlier diagnosis and initiation of disease modifying therapy.

Backpropagation-Free Multi-modal On-Device Model Adaptation via Cloud-Device Collaboration

In our increasingly interconnected world, where intelligent devices continually amass copious personalized multi-modal data, a pressing need arises to deliver high-quality, personalized device-aware services. However, this endeavor presents a multifaceted challenge to prevailing artificial intelligence (AI) systems primarily rooted in the cloud. As these systems grapple with shifting data distributions between the cloud and devices, the traditional approach of fine-tuning-based adaptation (FTA) exists the following issues: the costly and time-consuming data annotation required by FTA and the looming risk of model overfitting. To surmount these challenges, we introduce a Universal On-Device Multi-modal Model Adaptation Framework, revolutionizing on-device model adaptation by striking a balance between efficiency and effectiveness. The framework features the Fast Domain Adaptor (FDA) hosted in the cloud, providing tailored parameters for the Lightweight Multi-modal Model on devices. To enhance adaptability across multi-modal tasks, the AnchorFrame Distribution Reasoner (ADR) minimizes communication costs. Our contributions, encapsulated in the Cloud-Device Collaboration Multi-modal Parameter Generation (CDC-MMPG) framework, represent a pioneering solution for on-Device Multi-modal Model Adaptation (DMMA). Extensive experiments validate the efficiency and effectiveness of our method, particularly in video question answering and retrieval tasks, driving forward the integration of intelligent devices into our daily lives.

Swin Transformer with Enhanced Dropout and Layer-wise Unfreezing for Facial Expression Recognition in Mental Health Detection

This study presents an improved Facial Expression Recognition (FER) model using Swin transformers for enhanced performance in detecting mental health through facial emotion analysis. In addition, some techniques involving better dropout and layer-wise unfreezing were implemented to reduce model overfitting. This study evaluates the proposed models on benchmark datasets such as FER2013 and CK+ and real-time Genius HR data. Model A has no dropout layer, Model B has focal loss, and Model C has enhanced dropout and layer-wise unfreezing. Model C was the best among all proposed models, achieving test accuracies of 71.23% on FER2013 and 78.65% on CK+. Weighted cross-entropy loss and image augmentation were used to handle class imbalance. Based on Model C emotion predictions, a scoring mechanism was designed to analyze employees' mental health for the next 30 days. The higher the score, the higher the risk of mental health. This study demonstrates a practical version of the Swin transformer in FER models for detecting and early mental health intervention.

Predicting adolescent psychopathology from early life factors: A machine learning tutorial

The successful implementation and interpretation of machine learning (ML) models in epidemiological studies can be challenging without an extensive programming background. We provide a didactic example of machine learning for risk prediction in this study by determining whether early life factors could be useful for predicting adolescent psychopathology. In total, 9643 adolescents ages 9-10 from the Adolescent Brain and Cognitive Development (ABCD) Study were included in ML analysis to predict high Child Behavior Checklist (CBCL) scores (i.e., t-scores ≥60). ML models were constructed using a series of predictor combinations (prenatal, family history, sociodemographic) across 5 different algorithms. We assessed ML performance through sensitivity, specificity, F1-score, and area under the curve (AUC) metrics. A total of 1267 adolescents (13.1%) were found to have high CBCL scores. The best performing algorithms were elastic net and gradient boosted trees. The best performing elastic net models included prenatal and family history factors (Sensitivity 0.654, Specificity 0.713; AUC 0.742, F1-score 0.401) and prenatal, family, history, and sociodemographic factors (Sensitivity 0.668, Specificity 0.704; AUC 0.745, F1-score 0.402). Across all 5 ML algorithms, family history factors (e.g., either parent had nervous breakdowns, trouble holding jobs/fights/police encounters, and counseling for mental issues) and sociodemographic covariates (e.g., maternal age, child's sex, caregiver income and caregiver education) tended to be better predictors of adolescent psychopathology. The most important prenatal predictors were unplanned pregnancy, birth complications, and pregnancy complications. Our results suggest that inclusion of prenatal, family history, and sociodemographic factors in ML models can generate moderately accurate predictions of adolescent psychopathology. Issues associated with model overfitting, hyperparameter tuning, and system seed setting should be considered throughout model training, testing, and validation. Future early risk predictions models may improve with the inclusion of additional relevant covariates.