3FS-CBR-IRF: improving case retrieval for case-based reasoning with three feature selection and improved random forest
Ilhem Tarchoune,Akila Djebbar,Hayet Farida Djellali Merouani,Djamel Zenakhra,Tarchoune, Ilhem
DOI: https://doi.org/10.1007/s11042-024-18360-3
IF: 2.577
2024-02-11
Multimedia Tools and Applications
Abstract:Case-based reasoning (CBR) is widely used in medical decision support systems because of its similarity to human reasoning. Despite the effectiveness of the latter, its performance can be improved when reinforced by other automatic learning techniques. This is what induced us to propose a hybrid model 3FS-CBR-IRF (Three feature selection- Case based reasoning- Improved random forest) that combines case-based reasoning and random forests. Firstly, a feature selection step is used by three techniques such as: Correlation, Dropping Constant Features and Chi-square to select the relevant features. Secondly, we have modeled the retrieval phase of the CBR system by the modified random forests; in this step two pruning methods (Pre-pruning and Post-pruning) are applied only to the highly ranked features to improve the performance of the CBR system. Thirdly, the adaptation phase of the retrieval cases based on the adaptation rules is proposed. The developed approach is evaluated on 13 medical databases using the precision and several other performance criteria, the results obtained are satisfactory with an average precision of 91% for the first model (CBR-FS-RF) which improves the CBR system by feature selection techniques with classic random forest, 97% for the model (CBR-FS-IRF) which improve the CBR system by feature selection techniques with improved random forests and 98% for the second model (CBR-FS-IRF) with adaptation. The results showed that the proposed pruning method selects the best tree in less time and with better precision than hybridization with classical random forest that means an improvement of the performance of the CBR system.
computer science, information systems, theory & methods,engineering, electrical & electronic, software engineering