Associative Classifier Coupled With Unsupervised Feature Reduction for Dengue Fever Classification Using Gene Expression Data

• Published in: IEEE access Vol. 10; pp. 88340 - 88353
• Main Authors: Sen, Diptaraj, Paladhi, Saubhik, Frnda, Jaroslav, Chatterjee, Sankhadeep, Banerjee, Soumen, Nedoma, Jan
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2022; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Apriori algorithm; association rules; Biology; Classification; Classification algorithms; Classifiers; Clustering; Clustering algorithms; Clustering methods; Data mining; Dengue fever; Discretization; FP-growth algorithm; Gene expression; Gene expression data; Itemsets; Viruses (medical)
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2022.3198937

Recent studies have established the potential of classifiers designed using association rule mining methods. The current study proposes such an associative classifier to efficiently detect dengue fever using gene expression data. Labelled gene expression data has been preprocessed and discretized to mine association rules using well-established rule mining methods. Thereafter, unsupervised clustering methods have been applied to the discretized gene expression data to reduce and select the most promising features. The final feature reduced discretized gene expression data is subsequently used to mine rules in order to classify subjects into 'Dengue Fever' or 'Healthy'. Two well-known association rule mining methods, viz., Apriori and FP-Growth, have been used here along with various types of well established clustering methods. Extensive analysis has been reported with performance parameters in terms of accuracy, precision, recall and false positive rate using 5-fold cross-validation. Furthermore, a separate investigation has been conducted to find the most suitable number of features and confidence of association rule mining methods. The experimental results obtained indicate accurate detection of dengue fever patients at an early stage using the proposed associative classification method.