Interpreting complex time series forecasting models is challenging due to the temporal dependencies between time steps and the dynamic relevance of input features over time. Existing interpretation methods are limited by focusing mostly on classification tasks, evaluating using custom baseline models instead of the latest time series models, using simple synthetic datasets, and requiring training another model. We introduce a novel interpretation method called Windowed Temporal Saliency Rescaling (WinTSR) addressing these limitations. WinTSR explicitly captures temporal dependencies among the past time steps and efficiently scales the feature importance with this time importance. We benchmark WinTSR against 10 recent interpretation techniques with 5 state-of-the-art deep-learning models of different architectures, including a time series foundation model. We use 3 real-world datasets for both time-series classification and regression. Our comprehensive analysis shows that WinTSR significantly outranks the other local interpretation methods in overall performance. Finally, we provide a novel and open-source framework to interpret the latest time series transformers and foundation models.
Integrated Gradients is a well-known technique for explaining deep learning models. It calculates feature importance scores by employing a gradient based approach computing gradients of the model output with respect to input features and accumulating them along a linear path. While this works well for continuous features spaces, it may not be the most optimal way to deal with discrete spaces like word embeddings. For interpreting LLMs (Large Language Models), there exists a need for a non-linear path where intermediate points, whose gradients are to be computed, lie close to actual words in the embedding space. In this paper, we propose a method called Uniform Discretized Integrated Gradients (UDIG) based on a new interpolation strategy where we choose a favorable nonlinear path for computing attribution scores suitable for predictive language models. We evaluate our method on two types of NLP tasks- Sentiment Classification and Question Answering against three metrics viz Log odds, Comprehensiveness and Sufficiency. For sentiment classification, we have used the SST2, IMDb and Rotten Tomatoes datasets for benchmarking and for Question Answering, we have used the fine-tuned BERT model on SQuAD dataset. Our approach outperforms the existing methods in almost all the metrics.
This study investigates the dynamic relationship between the built environment and travel in Austin, Texas, over a 20-year period. Using three waves of household travel surveys from 1997, 2006, and 2017, the research employs a repeated cross-sectional approach to address the limitations of traditional longitudinal and cross-sectional studies. Methodologically, it integrates machine learning and inferential modeling to uncover non-linear relationships and threshold effects of built environment characteristics on travel. Findings reveal that the built environment serves as a sustainable tool for managing travel in the long term, contributing 50% or more to the total feature importance in predicting individual travel-surpassing the combined effects of personal and household characteristics. Increased transit accessibility, local and regional destination accessibility, population and employment density, and diversity significantly reduce travel, particularly within their identified thresholds, though the magnitude of their influence varies across time periods. These findings highlight the potential of smart growth policies-such as expanding transit accessibility, promoting high-density and mixed-use development, and discouraging single-use development and peripheral sprawl-as effective strategies to reduce car dependency and manage travel demand.
This study investigates crash severity risk modeling strategies for work zones involving large vehicles (i.e., trucks, buses, and vans) under crash data imbalance between low-severity (LS) and high-severity (HS) crashes. We utilized crash data involving large vehicles in South Carolina work zones from 2014 to 2018, which included four times more LS crashes than HS crashes. The objective of this study is to evaluate the crash severity prediction performance of various statistical, machine learning, and deep learning models under different feature selection and data balancing techniques. Findings highlight a disparity in LS and HS predictions, with lower accuracy for HS crashes due to class imbalance and feature overlap. Discriminative Mutual Information (DMI) yields the most effective feature set for predicting HS crashes without requiring data balancing, particularly when paired with gradient boosting models and deep neural networks such as CatBoost, NeuralNetTorch, XGBoost, and LightGBM. Data balancing techniques such as NearMiss-1 maximize HS recall when combined with DMI-selected features and certain models such as LightGBM, making them well-suited for HS crash prediction. Conversely, RandomUnderSampler, HS Class Weighting, and RandomOverSampler achieve more balanced performance, which is defined as an equitable trade-off between LS and HS metrics, especially when applied to NeuralNetTorch, NeuralNetFastAI, CatBoost, LightGBM, and Bayesian Mixed Logit (BML) using merged feature sets or models without feature selection. The insights from this study offer safety analysts guidance on selecting models, feature selection, and data balancing techniques aligned with specific safety goals, providing a robust foundation for enhancing work-zone crash severity prediction.
Accurate diagnosis of gait impairments is often hindered by subjective or costly assessment methods, with current solutions requiring either expensive multi-camera equipment or relying on subjective clinical observation. There is a critical need for accessible, objective tools that can aid in gait assessment while preserving patient privacy. In this work, we present a mobile phone-based, privacy-preserving artificial intelligence (AI) system for classifying gait impairments and introduce a novel dataset of 743 videos capturing seven distinct gait patterns. The dataset consists of frontal and sagittal views of trained subjects simulating normal gait and six types of pathological gait (circumduction, Trendelenburg, antalgic, crouch, Parkinsonian, and vaulting), recorded using standard mobile phone cameras. Our system achieved 86.5% accuracy using combined frontal and sagittal views, with sagittal views generally outperforming frontal views except for specific gait patterns like Circumduction. Model feature importance analysis revealed that frequency-domain features and entropy measures were critical for classifcation performance, specifically lower limb keypoints proved most important for classification, aligning with clinical understanding of gait assessment. These findings demonstrate that mobile phone-based systems can effectively classify diverse gait patterns while preserving privacy through on-device processing. The high accuracy achieved using simulated gait data suggests their potential for rapid prototyping of gait analysis systems, though clinical validation with patient data remains necessary. This work represents a significant step toward accessible, objective gait assessment tools for clinical, community, and tele-rehabilitation settings
Creativity is a fundamental skill of human cognition. We use textual forma mentis networks (TFMN) to extract network (semantic/syntactic associations) and emotional features from approximately one thousand human- and GPT3.5-generated stories. Using Explainable Artificial Intelligence (XAI), we test whether features relative to Mednick's associative theory of creativity can explain creativity ratings assigned by humans and GPT-3.5. Using XGBoost, we examine three scenarios: (i) human ratings of human stories, (ii) GPT-3.5 ratings of human stories, and (iii) GPT-3.5 ratings of GPT-generated stories. Our findings reveal that GPT-3.5 ratings differ significantly from human ratings not only in terms of correlations but also because of feature patterns identified with XAI methods. GPT-3.5 favours 'its own' stories and rates human stories differently from humans. Feature importance analysis with SHAP scores shows that: (i) network features are more predictive for human creativity ratings but also for GPT-3.5's ratings of human stories; (ii) emotional features played a greater role than semantic/syntactic network structure in GPT-3.5 rating its own stories. These quantitative results underscore key limitations in GPT-3.5's ability to align with human assessments of creativity. We emphasise the need for caution when using GPT-3.5 to assess and generate creative content, as it does not yet capture the nuanced complexity that characterises human creativity.
Deep learning has achieved remarkable success in processing and managing unstructured data. However, its "black box" nature imposes significant limitations, particularly in sensitive application domains. While existing interpretable machine learning methods address some of these issues, they often fail to adequately consider feature correlations and provide insufficient evaluation of model decision paths. To overcome these challenges, this paper introduces Real Explainer (RealExp), an interpretability computation method that decouples the Shapley Value into individual feature importance and feature correlation importance. By incorporating feature similarity computations, RealExp enhances interpretability by precisely quantifying both individual feature contributions and their interactions, leading to more reliable and nuanced explanations. Additionally, this paper proposes a novel interpretability evaluation criterion focused on elucidating the decision paths of deep learning models, going beyond traditional accuracy-based metrics. Experimental validations on two unstructured data tasks -- image classification and text sentiment analysis -- demonstrate that RealExp significantly outperforms existing methods in interpretability. Case studies further illustrate its practical value: in image classification, RealExp aids in selecting suitable pre-trained models for specific tasks from an interpretability perspective; in text classification, it enables the optimization of models and approximates the performance of a fine-tuned GPT-Ada model using traditional bag-of-words approaches.
With the growing demand for interpretable deep learning models, this paper introduces Integrative CAM, an advanced Class Activation Mapping (CAM) technique aimed at providing a holistic view of feature importance across Convolutional Neural Networks (CNNs). Traditional gradient-based CAM methods, such as Grad-CAM and Grad-CAM++, primarily use final layer activations to highlight regions of interest, often neglecting critical features derived from intermediate layers. Integrative CAM addresses this limitation by fusing insights across all network layers, leveraging both gradient and activation scores to adaptively weight layer contributions, thus yielding a comprehensive interpretation of the model's internal representation. Our approach includes a novel bias term in the saliency map calculation, a factor frequently omitted in existing CAM techniques, but essential for capturing a more complete feature importance landscape, as modern CNNs rely on both weighted activations and biases to make predictions. Additionally, we generalize the alpha term from Grad-CAM++ to apply to any smooth function, expanding CAM applicability across a wider range of models. Through extensive experiments on diverse and complex datasets, Integrative CAM demonstrates superior fidelity in feature importance mapping, effectively enhancing interpretability for intricate fusion scenarios and complex decision-making tasks. By advancing interpretability methods to capture multi-layered model insights, Integrative CAM provides a valuable tool for fusion-driven applications, promoting the trustworthy and insightful deployment of deep learning models.
Emotion recognition is significantly enhanced by integrating multimodal biosignals and IMU data from multiple domains. In this paper, we introduce a novel multi-scale attention-based LSTM architecture, combined with Squeeze-and-Excitation (SE) blocks, by leveraging multi-domain signals from the head (Meta Quest Pro VR headset), trunk (Equivital Vest), and peripheral (Empatica Embrace Plus) during affect elicitation via visual stimuli. Signals from 23 participants were recorded, alongside self-assessed valence and arousal ratings after each stimulus. LSTM layers extract features from each modality, while multi-scale attention captures fine-grained temporal dependencies, and SE blocks recalibrate feature importance prior to classification. We assess which domain's signals carry the most distinctive emotional information during VR experiences, identifying key biosignals contributing to emotion detection. The proposed architecture, validated in a user study, demonstrates superior performance in classifying valance and arousal level (high / low), showcasing the efficacy of multi-domain and multi-modal fusion with biosignals (e.g., TEMP, EDA) with IMU data (e.g., accelerometer) for emotion recognition in real-world applications.
Multi-modal large language models (MLLMs), such as GPT-4o, excel at integrating text and visual data but face systematic challenges when interpreting ambiguous or incomplete visual stimuli. This study leverages statistical modeling to analyze the factors driving these errors, using a dataset of geometric stimuli characterized by features like 3D, rotation, and missing face/side. We applied parametric methods, non-parametric methods, and ensemble techniques to predict classification errors, with the non-linear gradient boosting model achieving the highest performance (AUC=0.85) during cross-validation. Feature importance analysis highlighted difficulties in depth perception and reconstructing incomplete structures as key contributors to misclassification. These findings demonstrate the effectiveness of statistical approaches for uncovering limitations in MLLMs and offer actionable insights for enhancing model architectures by integrating contextual reasoning mechanisms.
A counterattack in soccer is a high speed, high intensity direct attack that can occur when a team transitions from a defensive state to an attacking state after regaining possession of the ball. The aim is to create a goal-scoring opportunity by convering a lot of ground with minimal passes before the opposing team can recover their defensive shape. The purpose of this research is to build gender-specific Graph Neural Networks to model the likelihood of a counterattack being successful and uncover what factors make them successful in professional soccer. These models are trained on a total of 20863 frames of synchronized on-ball event and spatiotemporal (broadcast) tracking data. This dataset is derived from 632 games of MLS (2022), NWSL (2022) and international soccer (2020-2022). With this data we demonstrate that gender-specific Graph Neural Networks outperform architecturally identical gender-ambiguous models in predicting the successful outcome of counterattacks. We show, using Permutation Feature Importance, that byline to byline speed, angle to the goal, angle to the ball and sideline to sideline speed are the node features with the highest impact on model performance. Additionally, we offer some illustrative examples on how to navigate the infinite solution search space to aid in identifying improvements for player decision making. This research is accompanied by an open-source repository containing all data and code, and it is also accompanied by an open-source Python package which simplifies converting spatiotemporal data into graphs. This package also facilitates testing, validation, training and prediction with this data. This should allow the reader to replicate and improve upon our research more easily.
In modern physical education, data-driven evaluation methods have gradually attracted attention, especially the quantitative prediction of students' sports performance through machine learning model. The purpose of this study is to use a variety of machine learning models to regress and predict students' comprehensive scores in baseball training, so as to evaluate the effectiveness of the current baseball teaching methods and put forward targeted training optimization suggestions. We set up a model and evaluate the performance of students by collecting many characteristics, such as hitting times, running times and batting. The experimental results show that K-Neighbors Regressor and Gradient Boosting Regressor are excellent in comprehensive prediction accuracy and stability, and the R score and error index are significantly better than other models. In addition, through the analysis of feature importance, it is found that cumulative hits and cumulative runs are the key factors affecting students' comprehensive scores. Based on the results of this study, this paper puts forward some suggestions on optimizing training strategies to help students get better performance in baseball training. The results show that the data-driven teaching evaluation method can effectively support physical education and promote personalized and refined teaching plan design.
We introduce CatNet, an algorithm that effectively controls False Discovery Rate (FDR) and selects significant features in LSTM. CatNet employs the derivative of SHAP values to quantify the feature importance, and constructs a vector-formed mirror statistic for FDR control with the Gaussian Mirror algorithm. To avoid instability due to nonlinear or temporal correlations among features, we also propose a new kernel-based independence measure. CatNet performs robustly on different model settings with both simulated and real-world data, which reduces overfitting and improves interpretability of the model. Our framework that introduces SHAP for feature importance in FDR control algorithms and improves Gaussian Mirror can be naturally extended to other time-series or sequential deep learning models.
This study pioneers the application of Recursive Feature Machines (RFM) in QSPR modeling, introducing a tailored feature importance analysis approach to enhance interpretability. By leveraging deep feature learning through AGOP, RFM achieves state-of-the-art (SOTA) results in predicting molecular properties, as demonstrated through solubility prediction across nine benchmark datasets. To capture a wide array of structural information, we employ diverse molecular representations, including MACCS keys, Morgan fingerprints, and a custom multi-scale hybrid fingerprint (HF) derived from global descriptors and SMILES local fragmentation techniques. Notably, the HF offers significant advantages over MACCS and Morgan fingerprints in revealing structural determinants of molecular properties. The feature importance analysis in RFM provides robust local and global explanations, effectively identifying structural features that drive molecular behavior and offering valuable insights for drug development. Additionally, RFM demonstrates strong redundancy-filtering abilities, as model performance remains stable even after removing redundant features within custom fingerprints. Importantly, RFM introduces the deep feature learning capabilities of the average gradient outer product (AGOP) matrix into ultra-fast kernel machine learning, to imbue kernel machines with interpretable deep feature learning capabilities. We extend this approach beyond the Laplace Kernel to the Matern, Rational Quadratic, and Gaussian kernels, to find that the Matern and Laplace kernels deliver the best performance, thus reinforcing the flexibility and effectiveness of AGOP in RFM. Experimental results show that RFM-HF surpasses both traditional machine learning models and advanced graph neural networks.
We investigate the effectiveness of Explainable AI (XAI) in verifying Machine Unlearning (MU) within the context of harbor front monitoring, focusing on data privacy and regulatory compliance. With the increasing need to adhere to privacy legislation such as the General Data Protection Regulation (GDPR), traditional methods of retraining ML models for data deletions prove impractical due to their complexity and resource demands. MU offers a solution by enabling models to selectively forget specific learned patterns without full retraining. We explore various removal techniques, including data relabeling, and model perturbation. Then, we leverage attribution-based XAI to discuss the effects of unlearning on model performance. Our proof-of-concept introduces feature importance as an innovative verification step for MU, expanding beyond traditional metrics and demonstrating techniques' ability to reduce reliance on undesired patterns. Additionally, we propose two novel XAI-based metrics, Heatmap Coverage (HC) and Attention Shift (AS), to evaluate the effectiveness of these methods. This approach not only highlights how XAI can complement MU by providing effective verification, but also sets the stage for future research to enhance their joint integration.
Data breaches have begun to take on new dimensions and their prediction is becoming of great importance to organizations. Prior work has addressed this issue mainly from a technical perspective and neglected other interfering aspects such as the social media dimension. To fill this gap, we propose STRisk which is a predictive system where we expand the scope of the prediction task by bringing into play the social media dimension. We study over 3800 US organizations including both victim and non-victim organizations. For each organization, we design a profile composed of a variety of externally measured technical indicators and social factors. In addition, to account for unreported incidents, we consider the non-victim sample to be noisy and propose a noise correction approach to correct mislabeled organizations. We then build several machine learning models to predict whether an organization is exposed to experience a hacking breach. By exploiting both technical and social features, we achieve a Area Under Curve (AUC) score exceeding 98%, which is 12% higher than the AUC achieved using only technical features. Furthermore, our feature importance analysis reveals that open ports and expired certificates are the best technical predictors, while spreadability and agreeability are the best social predictors.
Efforts are needed to identify and measure both communities' exposure to climate hazards and the social vulnerabilities that interact with these hazards, but the science of validating hazard vulnerability indicators is still in its infancy. Progress is needed to improve: 1) the selection of variables that are used as proxies to represent hazard vulnerability; 2) the applicability and scale for which these indicators are intended, including their transnational applicability. We administered an international urban survey in Buenos Aires, Argentina; Johannesburg, South Africa; London, United Kingdom; New York City, United States; and Seoul, South Korea in order to collect data on exposure to various types of extreme weather events, socioeconomic characteristics commonly used as proxies for vulnerability (i.e., income, education level, gender, and age), and additional characteristics not often included in existing composite indices (i.e., queer identity, disability identity, non-dominant primary language, and self-perceptions of both discrimination and vulnerability to flood risk). We then use feature importance analysis with gradient-boosted decision trees to measure the importance that these variables have in predicting exposure to various types of extreme weather events. Our results show that non-traditional variables were more relevant to self-reported exposure to extreme weather events than traditionally employed variables such as income or age. Furthermore, differences in variable relevance across different types of hazards and across urban contexts suggest that vulnerability indicators need to be fit to context and should not be used in a one-size-fits-all fashion.
Chronic kidney disease (CKD) is a significant public health challenge, often progressing to end-stage renal disease (ESRD) if not detected and managed early. Early intervention, warranted by silent disease progression, can significantly reduce associated morbidity, mortality, and financial burden. In this study, we propose a novel approach to modeling CKD progression using a combination of machine learning techniques and classical statistical models. Building on the work of Liu et al. (2023), we evaluate linear models, tree-based methods, and deep learning models to extract novel predictors for CKD progression, with feature importance assessed using Shapley values. These newly identified predictors, integrated with established clinical features from the Kidney Failure Risk Equation, are then applied within the framework of Cox proportional hazards models to predict CKD progression.
Birds exhibit a variety of flight styles, primarily classified as flapping, which is characterized by rapid up-and-down wing movements, and soaring, which involves gliding with wings outstretched. Each species usually performs specific flight styles, and this has been argued in terms of morphological and physiological adaptation. However, it remains a challenge to evaluate the contribution of each factor to the difference in flight styles. In this study, using phenotypic data from 635 migratory bird species, such as body mass, wing length, and breeding periods, we quantified the relative importance of each feature using Feature Importance and SHAP values, and used them to construct weighted L1 distance matrices and construct NJ trees. Comparison with traditional phylogenetic logistic regression revealed similarity in top-ranked features, but also differences in overall weight distributions and clustering patterns in NJ trees. Our results highlight the complexity of constructing a biologically useful distance matrix from correlated phenotypic features, while the complementary nature of these weighting methods suggests the potential utility of multi-faceted approaches to assessing feature contributions.
This study investigates the correlation between dairy farm characteristics and methane concentrations as derived from satellite observations in Eastern Canada. Utilizing data from 11 dairy farms collected between January 2020 and December 2022, we integrated Sentinel-5P satellite methane data with critical farm-level attributes, including herd genetics, feeding practices, and management strategies. Initial analyses revealed significant correlations with methane concentrations, leading to the application of Variance Inflation Factor (VIF) and Principal Component Analysis (PCA) to address multicollinearity and enhance model stability. Subsequently, machine learning models - specifically Random Forest and Neural Networks - were employed to evaluate feature importance and predict methane emissions. Our findings indicate a strong negative correlation between the Estimated Breeding Value (EBV) for protein percentage and methane concentrations, suggesting that genetic selection for higher milk protein content could be an effective strategy for emissions reduction. The integration of atmospheric transport models with satellite data further refined our emission estimates, significantly enhancing accuracy and spatial resolution. This research underscores the potential of advanced satellite monitoring, machine learning techniques, and atmospheric modeling in improving methane emission assessments within the dairy sector. It emphasizes the critical role of farm-specific characteristics in developing effective mitigation strategies. Future investigations should focus on expanding the dataset and incorporating inversion modeling for more precise emission quantification. Balancing ecological impacts with economic viability will be essential for fostering sustainable dairy farming practices.