https://www.mdu.se/

In the pursuit of reducing traffic accidents, drivers' mental workload (MWL) has been considered as one of the vital aspects. To measure MWL in different driving situations Electroencephalography (EEG) of the drivers has been studied intensely. However, in the literature, mostly, manual analytic methods are applied to extract and select features from the EEG signals to quantify drivers' MWL. Nevertheless, the amount of time and effort required to perform prevailing feature extraction techniques leverage the need for automated feature extraction techniques. This work investigates deep learning (DL) algorithm to extract and select features from the EEG signals during naturalistic driving situations. Here, to compare the DL based and traditional feature extraction techniques, a number of classifiers have been deployed. Results have shown that the highest value of area under the curve of the receiver operating characteristic (AUC-ROC) is 0.94, achieved using the features extracted by CNN-AE and support vector machine. Whereas, using the features extracted by the traditional method, the highest value of AUC-ROC is 0.78 with the multi-layer perceptron. Thus, the outcome of this study shows that the automatic feature extraction techniques based on CNN-AE can outperform the manual techniques in terms of classification accuracy.

Analysis of physiological signals, electroencephalography in more specific notion, is considered as a very promising technique to obtain objective measures for mental workload evaluation, however, it requires complex apparatus to record and thus with poor usability in monitoring in-vehicle drivers’mental workload. This study proposes amethodology of constructing a novel mutual information-based feature set from the fusion of electroencephalography and vehicular signals acquired through real driving experiment and deployed in evaluating drivers’ mental workload. Mutual information of electroencephalography and vehicular signals were used as the prime factor for the fusion of features. In order to assess the reliability of the developed feature set mental workload score prediction, classification and event classification tasks were performed using different machine learning models. Moreover, features extracted from electroencephalography were used to compare the performance. In the prediction of mental workload score, expert-defined scores were used as the target values. For classification tasks, true labels were set from contextual information of the experiment. An extensive evaluation of every prediction tasks was carried out using different validation methods. In predicting mental workload score from the proposed feature set lowest mean absolute error was 0.09 and for classifying mental workload highest accuracy was 94%. According to the outcome of the study, it can be stated that the novel mutual information based features developed through the proposed approach can be employed to classify and monitor in-vehicle drivers’ mental workload.

Artificial intelligence (AI) and machine learning (ML) have recently been radically improved and are now being employed in almost every application domain to develop automated or semi-automated systems. To facilitate greater human acceptability of these systems, explainable artificial intelligence (XAI) has experienced significant growth over the last couple of years with the development of highly accurate models but with a paucity of explainability and interpretability. The literature shows evidence from numerous studies on the philosophy and methodologies of XAI. Nonetheless, there is an evident scarcity of secondary studies in connection with the application domains and tasks, let alone review studies following prescribed guidelines, that can enable researchers’ understanding of the current trends in XAI, which could lead to future research for domain- and application-specific method development. Therefore, this paper presents a systematic literature review (SLR) on the recent developments of XAI methods and evaluation metrics concerning different application domains and tasks. This study considers 137 articles published in recent years and identified through the prominent bibliographic databases. This systematic synthesis of research articles resulted in several analytical findings: XAI methods are mostly developed for safety-critical domains worldwide, deep learning and ensemble models are being exploited more than other types of AI/ML models, visual explanations are more acceptable to end-users and robust evaluation metrics are being developed to assess the quality of explanations. Research studies have been performed on the addition of explanations to widely used AI/ML models for expert users. However, more attention is required to generate explanations for general users from sensitive domains such as finance and the judicial system.

Numerous studies have exploited the potential of Artificial Intelligence (AI) and Machine Learning (ML) models to develop intelligent systems in diverse domains for complex tasks, such as analysing data, extracting features, prediction, recommendation etc. However, presently these systems embrace acceptability issues from the end-users. The models deployed at the back of the systems mostly analyse the correlations or dependencies between the input and output to uncover the important characteristics of the input features, but they lack explainability and interpretability that causing the acceptability issues of intelligent systems and raising the research domain of eXplainable Artificial Intelligence (XAI). In this study, to overcome these shortcomings, a hybrid XAI approach is developed to explain an AI/ML model's inference mechanism as well as the final outcome. The overall approach comprises of 1) a convolutional encoder that extracts deep features from the data and computes their relevancy with features extracted using domain knowledge, 2) a model for classifying data points using the features from autoencoder, and 3) a process of explaining the model's working procedure and decisions using mutual information to provide global and local interpretability. To demonstrate and validate the proposed approach, experimentation was performed using an electroencephalography dataset from road safety to classify drivers' in-vehicle mental workload. The outcome of the experiment was found to be promising that produced a Support Vector Machine classifier for mental workload with approximately 89% performance accuracy. Moreover, the proposed approach can also provide an explanation for the classifier model's behaviour and decisions with the combined illustration of Shapely values and mutual information.

Understanding individual car drivers’ behavioural variations and heterogeneity is a significant aspect of developingcar simulator technologies, which are widely used in transport safety. This also characterizes the heterogeneity in drivers’ behaviour in terms of risk and hurry, using both real-time on-track and in-simulator driving performance features. Machine learning (ML) interpretability has become increasingly crucial for identifying accurate and relevant structural relationships between spatial events and factors that explain drivers’ behaviour while being classified and the explanations for them are evaluated. However, the high predictive power of ML algorithms ignore the characteristics of non-stationary domain relationships in spatiotemporal data (e.g., dependence, heterogeneity), which can lead to incorrect interpretations and poor management decisions. This study addresses this critical issue of ‘interpretability’ in ML-based modelling of structural relationships between the events and corresponding features of the car drivers’ behavioural variations. In this work, an exploratory experiment is described that contains simulator and real driving concurrently with a goal to enhance the simulator technologies. Here, initially, with heterogeneous data, several analytic techniques for simulator bias in drivers’ behaviour have been explored. Afterwards, five different ML classifier models were developed to classify risk and hurry in drivers’ behaviour in real and simulator driving. Furthermore, two different feature attribution-based explanation models were developed to explain the decision from the classifiers. According to the results and observation, among the classifiers, Gradient Boosted Decision Trees performed best with a classification accuracy of 98.62%. After quantitative evaluation, among the feature attribution methods, the explanation from Shapley Additive Explanations (SHAP) was found to be more accurate. The use of different metrics for evaluating explanation methods and their outcome lay the path toward further research in enhancing the feature attribution methods.

Feature attribution is a class of explainable artificial intelligence (XAI) methods that produce the contributions of data features to a model's decision. There are multiple accounts stating that feature attribution methods produce inconsistent results and should always be evaluated. However, the existing body of literature on evaluation techniques is still immature with multiple proposed techniques and a lack of widely adopted methods, making it difficult to recognize the best approach for each circumstance. This article investigates an approach to creating synthetic data for regression that can be used to evaluate the results of feature attribution methods. From a real-world dataset, the proposed approach describes how to create synthetic data that preserves the patterns of the original data and enables comprehensive evaluation of XAI methods. This research also demonstrates how global and local feature attributions can be represented in the additive form of case-based reasoning as a benchmark method for evaluation. Finally, this work demonstrates the case where a method that includes a standardization step does not produce feature attributions of the same quality as one that does not use standardization in the context of a regression task.

Tree-ensemble models, such as Extreme Gradient Boosting (XGBoost), are renowned Machine Learning models which have higher prediction accuracy compared to traditional tree-based models. This higher accuracy, however, comes at the cost of reduced interpretability. Also, the decision path or prediction rule of XGBoost is not explicit like the tree-based models. This paper proposes the iXGB--interpretable XGBoost, an approach to improve the interpretability of XGBoost. iXGB approximates a set of rules from the internal structure of XGBoost and the characteristics of the data. In addition, iXGB generates a set of counterfactuals from the neighbourhood of the test instances to support the understanding of the end-users on their operational relevance. The performance of iXGB in generating rule sets is evaluated with experiments on real and benchmark datasets which demonstrated reasonable interpretability. The evaluation result also supports that the interpretability of XGBoost can be improved without using surrogate methods.