Identifying dyslexia in school pupils from eye movement and demographic data using artificial intelligence.

This paper represents our research results in the pursuit of the following objectives: (i) to introduce a novel multi-sources data set to tackle the shortcomings of the previous data sets, (ii) to propose a robust artificial intelligence-based solution to identify dyslexia in primary school pupils, (iii) to investigate our psycholinguistic knowledge by studying the importance of the features in identifying dyslexia by our best AI model. In order to achieve the first objective, we collected and annotated a new set of eye-movement-during-reading data. Furthermore, we collected demographic data, including the measure of non-verbal intelligence, to form our three data sources. Our data set is the largest eye-movement data set globally. Unlike the previously introduced binary-class data sets, it contains (A) three class labels and (B) reading speed. Concerning the second objective, we formulated the task of dyslexia prediction as regression and classification problems and scrutinized the performance of 12 classifications and eight regressions approaches. We exploited the Bayesian optimization method to fine-tune the hyperparameters of the models: and reported the average and the standard deviation of our evaluation metrics in a stratified ten-fold cross-validation. Our studies showed that multi-layer perceptron, random forest, gradient boosting, and k-nearest neighbor form the group having the most acceptable results. Moreover, we showed that although separately using each data source did not lead to accurate results, their combination led to a reliable solution. We also determined the importance of the features of our best classifier: our findings showed that the IQ, gender, and age are the top three important features; we also showed that fixation along the y-axis is more important than other fixation data. Dyslexia detection, eye fixation, eye movement, demographic, classification, regression, artificial intelligence.

Community Partitioning over Feature-Rich Networks Using an Extended K-Means Method.

This paper proposes a meaningful and effective extension of the celebrated K-means algorithm to detect communities in feature-rich networks, due to our assumption of non-summability mode. We least-squares approximate given matrices of inter-node links and feature values, leading to a straightforward extension of the conventional K-means clustering method as an alternating minimization strategy for the criterion. This works in a two-fold space, embracing both the network nodes and features. The metric used is a weighted sum of the squared Euclidean distances in the feature and network spaces. To tackle the so-called curse of dimensionality, we extend this to a version that uses the cosine distances between entities and centers. One more version of our method is based on the Manhattan distance metric. We conduct computational experiments to test our method and compare its performances with those by competing popular algorithms at synthetic and real-world datasets. The cosine-based version of the extended K-means typically wins at the high-dimension real-world datasets. In contrast, the Manhattan-based version wins at most synthetic datasets.

Least-squares community extraction in feature-rich networks using similarity data.

We explore a doubly-greedy approach to the issue of community detection in feature-rich networks. According to this approach, both the network and feature data are straightforwardly recovered from the underlying unknown non-overlapping communities, supplied with a center in the feature space and intensity weight(s) over the network each. Our least-squares additive criterion allows us to search for communities one-by-one and to find each community by adding entities one by one. A focus of this paper is that the feature-space data part is converted into a similarity matrix format. The similarity/link values can be used in either of two modes: (a) as measured in the same scale so that one may can meaningfully compare and sum similarity values across the entire similarity matrix (summability mode), and (b) similarity values in one column should not be compared with the values in other columns (nonsummability mode). The two input matrices and two modes lead us to developing four different Iterative Community Extraction from Similarity data (ICESi) algorithms, which determine the number of communities automatically. Our experiments at real-world and synthetic datasets show that these algorithms are valid and competitive.