Instance-based Models Research Articles

Knowledge graph driven medicine recommendation system using graph neural networks on longitudinal medical records

Medicine recommendation systems are designed to aid healthcare professionals by analysing a patient’s admission data to recommend safe and effective medications. These systems are categorised into two types: instance-based and longitudinal-based. Instance-based models only consider the current admission, while longitudinal models consider the patient’s medical history. Electronic Health Records are used to incorporate medical history into longitudinal models. This project proposes a novel Knowledge Graph-Driven Medicine Recommendation System using Graph Neural Networks, KGDNet, that utilises longitudinal EHR data along with ontologies and Drug-Drug Interaction knowledge to construct admission-wise clinical and medicine Knowledge Graphs for every patient. Recurrent Neural Networks are employed to model a patient’s historical data, and Graph Neural Networks are used to learn embeddings from the Knowledge Graphs. A Transformer-based Attention mechanism is then used to generate medication recommendations for the patient, considering their current clinical state, medication history, and joint medical records. The model is evaluated on the MIMIC-IV EHR data and outperforms existing methods in terms of precision, recall, F1 score, Jaccard score, and Drug-Drug Interaction control. An ablation study on our models various inputs and components to provide evidence for the importance of each component in providing the best performance. Case study is also performed to demonstrate the real-world effectiveness of KGDNet.

Multiple instance learning for medical image classification based on instance importance

Multiple Instance Learning (MIL) is designed to classify instances where class labels are associated with sets of instances, a common occurrence in biomedical data, especially when multiple images are derived from a single object measurement. Among the various MIL techniques, instance-based classification emerges as the main approach. Individual instances are assigned class labels, and the final bag-level classification is determined through an aggregation process, usually based on average or maximum values. While instance-based models may exhibit slightly lower performance compared to more general methods, they offer the unique advantage of interpretability, allowing the decision-making process to be traced to each individual instance. This article presents an innovative instance-based approach that not only assigns class labels to instances, but also gives them meaning. The final aggregation uses a weighted average, allowing the model to identify instances carrying valuable information while distinguishing those that are redundant in the classification process. This enhancement gives the model the ability to recognize the meaning of individual instances, thereby increasing both its overall interpretability and classification performance.

An instance-based model account of the benefits of varied practice in visuomotor skill

Exposing learners to variability during training has been demonstrated to improve performance in subsequent transfer testing. Such variability benefits are often accounted for by assuming that learners are developing some general task schema or structure. However much of this research has neglected to account for differences in similarity between varied and constant training conditions. In a between-groups manipulation, we trained participants on a simple projectile launching task, with either varied or constant conditions. We replicate previous findings showing a transfer advantage of varied over constant training. Furthermore, we show that a standard similarity model is insufficient to account for the benefits of variation, but, if the model is adjusted to assume that varied learners are tuned towards a broader generalization gradient, then a similarity-based model is sufficient to explain the observed benefits of variation. Our results therefore suggest that some variability benefits can be accommodated within instance-based models without positing the learning of some schemata or structure.

Instance-Based Neural Dependency Parsing

AbstractInterpretable rationales for model predictions are crucial in practical applications. We develop neural models that possess an interpretable inference process for dependency parsing. Our models adopt instance-based inference, where dependency edges are extracted and labeled by comparing them to edges in a training set. The training edges are explicitly used for the predictions; thus, it is easy to grasp the contribution of each edge to the predictions. Our experiments show that our instance-based models achieve competitive accuracy with standard neural models and have the reasonable plausibility of instance-based explanations.

Patient-Provider Communication Training Models for Interactive Speech Devices.

Patient-provider communication plays a major role in healthcare with its main goal being to improve the patient's health and build a trustworthy relationship between the patient and the doctor. Provider's efficiency and effectiveness in communication can be improved through training in order to meet the essential elements of communication that are relevant during medical encounters. We surmised that speech-enabled conversational agents could be used as a training tool. In this study, we propose designing an ontology-based interaction model that can direct software agents to train dental and medical students. We transformed sample scenario scripts into a formalized ontology training model that links utterances of the user and the machine that expresses patient-provider communication. We created two instance-based models from the ontology to test the operational execution of the model using a prototype software engine. The assessment revealed that the dialogue engine was able to handle about 62% of the dialogue links. Future direction of this work will focus on further enhancing and capturing the features of patient-provider communication, and eventual deployment for pilot testing.

Machine Learning Methods to Analyze Injury Severity of Drivers from Different Age and Gender Groups

Access to non-biased and accurate models capable of predicting driver injury severity of collision events is vital for determining what safety measures should be implemented at intersections. Inadequate models can underestimate the potential for collision events to result in driver fatalities or injuries, which can lead to improperly assessing the safety criteria of an intersection. This study investigates how injury severity differs between drivers of various ages and gender groups using cost-sensitive data-mining models. Previous research efforts have used machine learning methods for predicting injury severity; however, these studies did not consider the consequences (cost) of incorrect predictions. This paper addresses this shortfall by considering the monetary cost of incorrect injury severity predictions when developing C4.5, instance-based (IB), and random forest (RF) machine-learning models. One model of each method was developed for four distinct cohorts of drivers (i.e., younger males, younger females, older males, and older females). Each model considered a selection of driver, vehicular, road/traffic, environmental, and crash parameters for determining if they significantly influenced driver injury severity. A five-year period of two-vehicle crash data collected at signalized intersections in the metropolitan area of Miami, Florida was used in the models. Results indicated that cost-sensitive learning classifiers were superior to regular classifiers at accurately predicting injuries and fatalities of crashes. Among cost-sensitive models, RF outperformed C4.5 and IB models in predicting driver injury severity for four groups of drivers. The models displayed substantial differences in injury severity determinants across the age/gender cohorts.

Analysis of Gap Acceptance Behavior for Unprotected Right and Left Turning Maneuvers at Signalized Intersections using Data Mining Methods: A Driving Simulation Approach

Gap acceptance predictability has become a vital area of interest for traffic safety and operations due to its complexity and significance in understanding a population’s driving behavior. Recent studies have implemented statistical modeling techniques, such as binary logit model (BLM), to predict drivers’ gap acceptance behaviors. However, these models have inherent presumptions and pre-set correlations that, if contravened, can produce erroneous results. The use of non-parametric data mining techniques, such as decision trees, avoids these deficiencies, thus resulting in improvements to the predictive capability of the models. In this study, the feasibility of C4.5 decision trees, instance-based (IB), and random forest (RF) models for predicting drivers’ gap decisions was examined by comparing their results with BLM. To accomplish this objective, 66 study participants drove through ten driving simulation scenarios requiring the navigation of unprotected right and left-turning maneuvers at four-legged, signalized intersections. The data collected from these tests will provide means to directly compare and rank the data mining and statistical models, while also allowing for the identification of variables that are significantly influencing gap acceptance. Results produced from the models indicated that data mining models were superior to BLM at accurately predicting a participant’s gap decisions. RF models outperformed the C4.5 and IB models in predicting gap acceptance behaviors for both the left and right turning scenarios. Because of its superior performance, the authors recommend the implementation of the RF model for predicting gap decisions at unprotected signalized intersections.

Source code size estimation approaches for object-oriented systems from UML class diagrams: A comparative study

BackgroundSource code size in terms of SLOC (source lines of code) is the input of many parametric software effort estimation models. However, it is unavailable at the early phase of software development. ObjectiveWe investigate the accuracy of early SLOC estimation approaches for an object-oriented system using the information collected from its UML class diagram available at the early software development phase. MethodWe use different modeling techniques to build the prediction models for investigating the accuracy of six types of metrics to estimate SLOC. The used techniques include linear models, non-linear models, rule/tree-based models, and instance-based models. The investigated metrics are class diagram metrics, predictive object points, object-oriented project size metric, fast&&serious class points, objective class points, and object-oriented function points. ResultsBased on 100 open-source Java systems, we find that the prediction model built using object-oriented project size metric and ordinary least square regression with a logarithmic transformation achieves the highest accuracy (mean MMRE=0.19 and mean Pred(25)=0.74). ConclusionWe should use object-oriented project size metric and ordinary least square regression with a logarithmic transformation to build a simple, accurate, and comprehensible SLOC estimation model.

The development of intuitive knowledge classifier and the modeling of domain dependent data

Creating an efficient user knowledge model is a crucial task for web-based adaptive learning environments in different domains. It is often a challenge to determine exactly what type of domain dependent data will be stored and how it will be evaluated by a user modeling system. The most important disadvantage of these models is that they classify the knowledge of users without taking into account the weight differences among the domain dependent data of users. For this purpose, both the probabilistic and the instance-based models have been developed and commonly used in the user modeling systems. In this study a powerful, efficient and simple ‘Intuitive Knowledge Classifier’ method is proposed and presented to model the domain dependent data of users. A domain independent object model, the user modeling approach and the weight-tuning method are combined with instance-based classification algorithm to improve classification performances of well-known the Bayes and the k-nearest neighbor-based methods. The proposed knowledge classifier intuitively explores the optimum weight values of students’ features on their knowledge class first. Then it measures the distances among the students depending on their data and the values of weights. Finally, it uses the dissimilarities in the classification process to determine their knowledge class. The experimental studies have shown that the weighting of domain dependent data of students and combination of user modeling algorithms and population-based searching approach play an essential role in classifying performance of user modeling system. The proposed system improves the classification accuracy of instance-based user modeling approach for all distance metrics and different k-values.

On the use of data filtering techniques for credit risk prediction with instance-based models

Many techniques have been proposed for credit risk prediction, from statistical models to artificial intelligence methods. However, very few research efforts have been devoted to deal with the presence of noise and outliers in the training set, which may strongly affect the performance of the prediction model. Accordingly, the aim of the present paper is to systematically investigate whether the application of filtering algorithms leads to an increase in accuracy of instance-based classifiers in the context of credit risk assessment. The experimental results with 20 different algorithms and 8 credit databases show that the filtered sets perform significantly better than the non-preprocessed training sets when using the nearest neighbour decision rule. The experiments also allow to identify which techniques are most robust and accurate when confronted with noisy credit data.

Analogy and the dual-route model of morphology

Abstract Prasada and Pinker's (1993) subjects provided past tense forms of nonce verbs. The subject's willingness to provide irregular past tense forms correlated with the verb's phonological similarity to existing irregular English verbs. However, there was no correlation between the number of nonce verbs assigned regular inflection, and the verb's similarity to existing regular verbs. According to the dual-route model, this is expected since irregular items are stored in associative memory, while regular items take an allomorph of -ed by rule. A singleroute connectionist simulation failed to duplicate the subject's behavior on regular verbs. Two instance-based models were applied to the data: Analogical Modeling of Language and the Tilburg Memory Based Learner. Each model employed a similarity algorithm to determine the behavior of all regular and irregular items. Both models successfully mirrored the subject's responses. Therefore, the data are consistent with an instance-based single-route model of morphology.

Face Categorization, Graded Priming, and the Mediating Influences of Similarity

Two experiments are reported that contrast instance-based and abstraction-based models of categorization within a priming paradigm. In both experiments, target person photos were preceded by prime photos that varied in their degree of similarity to the target photo to test how inter-item similarity influences gender and race categorization efficiency of the target photo. Results consistently indicate that, the more similar the prime item is to the target item, the greater the facilitation of categorical processing. The findings are robust across race and gender question types, independent of perceiver group membership. The results support the conclusion that categorical judgments are at least partially mediated by specific item comparisons. The findings are discussed for their relevance to models of priming and for models of person perception and decision making.

Remembering facial configurations

Eight experiments are reported showing that subjects can remember rather subtle aspects of the configuration of facial features to which they have earlier been exposed. Subjects saw several slightly different configurations (formed by altering the relative placement of internal features of the face) of each of ten different faces, and they were asked to rate the apparent age and masculinity-femininity of each. Afterwards, subjects were asked to select from pairs of faces the configuration which was identical to one previously rated. Subjects responded strongly to the central or “prototypical” configuration of each studied face where this was included as one member of each test pair, whether or not it had been studied (Experiments 1, 2 and 4). Subjects were also quite accurate at recognising one of the previously encountered extremes of the series of configurations that had been rated (Experiment 3), but when unseen prototypes were paired with seen exemplars subjects' performance was at chance (Experiment 5). Prototype learning of face patterns was shown to be stronger than that for house patterns, though both classes of patterns were affected equally by inversion (Experiment 6). The final two experiments demonstrated that preferences for the prototype could be affected by instructions at study and by whether different exemplars of the same face were shown consecutively or distributed through the study series. The discussion examines the implications of these results for theories of the representation of faces and for instance-based models of memory.