Automated reading level classification model based on improved orbital pattern
Automatic reading level for detection and classification is a challenging problem in machine learning. A multilevel feature extraction-based self-organized model may be useful to overcome this hurdle without using deep learning, which requires an ultra-large sample size. In this work, a novel speech...
Gespeichert in:
Veröffentlicht in: | Multimedia tools and applications 2024-05, Vol.83 (17), p.52819-52840 |
---|---|
Hauptverfasser: | , , , , , , , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | Automatic reading level for detection and classification is a challenging problem in machine learning. A multilevel feature extraction-based self-organized model may be useful to overcome this hurdle without using deep learning, which requires an ultra-large sample size. In this work, a novel speech dataset was collected from 57 primary school students by reading a fixed paragraph, and experts labeled these speeches as good, moderate, or bad. We then developed a handcrafted, self-organized learning model. We constructed a novel method using a multilevel feature extraction method, termed improved orbital pattern (IOP) and wavelet packet decomposition (WPD). The proposed IOP generates textural features from the speeches and the used wavelet bands. These extracted features are input to neighborhood components analysis (NCA) to reduce feature dimension. Then the feature set is input to the support vector machine (SVM) classifier to obtain loss values. The output of ten feature vectors of the NCA and SVM classifiers are merged to provide the final feature vector. The most significant 512 features were selected using the NCA feature selection function. These 512 features are classified via the SVM classifier with tenfold cross-validation (CV) and leave-one-subject-out (LOSO) validation strategies. The proposed IOP and WPD-based model yielded an accuracy of 92.75% with a tenfold CV and a 76.18% accuracy using LOSO validation strategies in classifying bad, intermediate, and good reading levels. Our developed model is ready to be validated with more data before its actual usage in schools to aid the teachers. |
---|---|
ISSN: | 1573-7721 1380-7501 1573-7721 |
DOI: | 10.1007/s11042-023-17535-8 |