JASBO: Jaya Average Subtraction Based Optimization with Deep Learning Model for Multi-Classification of Infectious Disease from Unstructured Data
| Autor: | Vian Sabeeh, Ahmed Bahaaulddin A. Alwahhab, Ali Abdulmunim Ibrahim Al-kharaz |
|---|---|
| Jazyk: | Arabic<br />English |
| Rok vydání: | 2024 |
| Předmět: | |
| Zdroj: | Baghdad Science Journal, Vol 21, Iss 10 (2024) |
| Druh dokumentu: | article |
| ISSN: | 2078-8665 2411-7986 |
| DOI: | 10.21123/bsj.2024.9184 |
| Popis: | Infectious diseases have become an unavoidable big trouble in today's environment with a similar symptomatology that makes difficult of early detection and clear separation of infection. Hence, it is required to generate a new technique that best utilizes the various symptomatologies present in the illnesses for its multi-classification. Medical documents are considered an essential source for modern, invented, and robust analysis methods for accurate infection diagnoses. Accordingly, enriching medical text processing is beneficial in health informatics. In this research, proposed Jaya Average Subtraction Based Optimization (JASBO), which is enabled by Deep Learning (DL) is used to classify infectious diseases into many categories from unstructured data. Moreover, the DL model used is Infectious Disease Network (ID-Net) which combines Convolutional Neural Network (CNN) and Bidirectional-Long Short-Term Memory (Bi-LSTM). To specify the strange or discriminative words with BI-LSTM . JASBO algorithm used in the model to determine the size of the filter in the final classification network to detect the meaningful part of the text. The input text is given to the Tokenization layer in this case, where the tokens get formed and is forwarded to CNN. Additionally, character-based network features are extracted using Bi-LSTM model. Then, vector representation is concatenated with two separate character-level extractions from Bi-LSTM and CNN. Character level features are passed to the attention layer, which uses the Kumar-Hassebrook similarity measure to calculate the score function. Label of each word token is then predicted by the ID layer, at which layer size is found by JASBO. Here, JASBO combines Jaya algorithm with an Average and Subtraction-Based Optimizer (ASBO). The best performance of JASBO_ID-Net is analyzed with three performance metrics: accuracy with superior value of 91%, recall with high value of 88.7%, and F-measure with a superior value of 90%. |
| Databáze: | Directory of Open Access Journals |
| Externí odkaz: |
|