DT-GA-FLANN: Decision Tree-Based Genetically Optimized FLANN for Missing Value Imputation and Classification

In the current landscape, addressing missing values poses a significant challenge when dealing with real-world problems. In this work, a hybrid model, DT-GA-FLANN, is proposed to address the missing value problem and enhance classifier performance. The missing values are imputed using Classification and Regression Trees (CART), ensuring robust data preprocessing. For classification, a genetically optimized Functional Link Artificial Neural Network (FLANN) is used, chosen for its faster convergence and lower computational complexity compared to multi-layer neural networks. However, functional expansion in FLANN can increase architectural complexity, which is effectively mitigated using Genetic Algorithm (GA). GA is selected for its adaptability, parallel processing capabilities, and efficiency in complex optimization tasks. The FLANN parameters are iteratively fine-tuned to maximize performance. The performance of DT-GA-FLANN is validated through extensive experiments on four real-world and two synthetic datasets. It is compared against ten hybrid models, comprising four statistical and six machine learning-based hybrid models. Experimental analysis reveals that DT-GA-FLANN achieves an average improvement in accuracy, precision, recall, and F-measure by 3.7%, 4.3%, 3.1%, and 3.8%, respectively over machine learning models, and by 10.2%, 12.4%, 10.8%, and 11.6%, respectively over statistical models on real-world datasets. In case of synthetic datasets, it further improves accuracy, precision, recall, and F-measure by 4.2%, 5.1%, 4.7%, and 4.3%, respectively over machine learning models, and by 9.1%, 10.8%, 9.4%, and 9.9%, respectively over statistical models. These results confirm DT-GA-FLANN’s superior classification performance and generalization ability across diverse datasets. The experimental results show that the genetically optimized FLANN significantly outperforms other hybrid models across multiple performance metrics.