In an article published in the journal Scientific Reports, researchers from the University of Tuscia, Italy, proposed a machine learning (ML)-based classification model to provide customized support digital tools and learning strategies for students with dyslexia. A real case scenario where the predicted and the actual usefulness of the tools and strategies were compared was used to evaluate the performance of the model.
Study: Personalized Support for Dyslexic University Students: A Machine Learning Approach. Image credit: Stivog/Shutterstock
Background
Dyslexia is a learning disorder that affects reading skills, comprehension, and memory and can have negative consequences on academic achievement, self-esteem, and psychological well-being. ML, a subset of artificial intelligence (AI), enables machines to learn from historical data and instructions and make predictions or decisions. It can be trained on large datasets to find patterns, correlations, or classifications that are not easily recognized by humans.
ML has been widely applied to various fields and domains, such as medicine, education, engineering, finance, psychology, and social sciences. It can offer innovative solutions to diagnose and support individuals with dyslexia, by analyzing their cognitive and behavioral test results, brain images, eye movements, and neurological data. Although most previous studies explored the ability of ML to offer customized support tools and strategies to dyslexic children or primary school students, there is a lack of specific tools for university students with dyslexia, who face different and more complex challenges in their academic careers.
About the Research
In the present paper, the authors created an ML model to classify the most effective methodologies for supporting university students with dyslexia, based on the challenges encountered during their learning process. The prediction algorithm relies on supervised ML techniques, indicating that it learns from labeled data with known outputs.
Data collection involved a self-evaluation questionnaire designed by a team of psychologists and distributed to over 1200 university students with dyslexia. The questionnaire covered difficulties in reading, writing, arithmetic, comprehension, memorization, attention, and organization, along with inquiries about the utility of 17 digital tools and 22 learning strategies. Responses were transformed into scores, serving as input and output variables for the ML algorithms.
The study assessed the performance of different ML algorithms, including random forest (RF), logistic regression (LR), k-nearest neighbors (kNN), and support vector machines (SVM). Various configurations were tested for each algorithm, involving treatments such as considering input variables as either numeric or binary values, using different kernels or distance metrics, and applying boosting techniques to enhance performance. Additionally, the accuracy of each algorithm and setup was evaluated through cross-validation and weighted metrics, considering the imbalance of output classes.
Research Findings
The outcomes showed that the implemented ML algorithms were able to achieve high prediction accuracy for most of the tools and strategies, with an average of 90.4%. The best-performing algorithm and setup for each tool or strategy were selected to compose the final classification model, which was tested on a real-world case scenario with 43 dyslexic students. The comparison between the algorithm output and the student’s feedback confirmed that the novel model can successfully suggest personalized support methodologies for dyslexic students, based on the issues encountered.
The authors also analyzed the most useful tools and strategies, according to the questionnaire data. They found that 17 tools and 22 strategies were considered useful by most of the students, with different degrees of preference. Text-to-speech software, mind maps, digital calendars, and spell checkers are the most useful tools. Some of the most useful strategies were studying in a quiet environment, using colors and images, summarizing the main concepts, and asking for clarifications.
Applications
The proposed classification model has significant potential applications in the fields of education and psychology. It can assist students with dyslexia in identifying optimal support methodologies tailored to their specific needs and preferences, thereby enhancing their learning outcomes and self-esteem. Additionally, the model can help teachers, tutors, and psychologists deliver more personalized and effective interventions for dyslexic students while monitoring their progress and addressing difficulties. Furthermore, it can contribute to developing innovative digital tools and learning strategies by leveraging student feedback and suggestions.
Conclusion
In summary, a novel ML-based classification model was proposed in this study to create a personalized support system for university students with dyslexia, based on their specific difficulties and the usefulness of various digital tools and strategies. The paper showed that the ML model achieved a high accuracy in predicting the best support methodologies for each student and that the model was validated on a real-case scenario.
The researchers suggest some directions for future research, such as extending the model to other languages, other specific learning disorders (SLDs), or other educational levels, improving data collection and processing methods, and exploring other advanced ML techniques, such as deep learning neural networks or reinforcement learning. They also highlight the importance of involving dyslexic students in the design and evaluation of support methods and promoting their awareness and empowerment.
Machine Learning Identifies Seismic Precursors, Advancing Earthquake Forecasting Capabilities