Handling High-Dimensional Data and Classification Using a Hybrid Feature Selection Approach
Shobit Agrawal, Dilip Kumar Sharma · 2022 2nd International Conference on Advance Computing and Innovative Technologies in Engineering (ICACITE) · 2022
Recently, we published a paper on feature selection and extraction technique on time series data classification: A Comparative analysis mainly focusing on classifiers, in this paper, we are implementing 10 classifiers on 3 open source datasets of UCI machine learning and measuring accuracy, precision, recall, and f1 score. On classifying using 10 classifiers over 3 open-source datasets results reveal that the confusion matrix attributes show the best results in the case of two classifiers, i.e. XGBoost and LightGBM. We proposed a hybrid approach by combining the filter class Minimum Redundancy and Maximum Relevance method and wrapper class Recursive Feature Elimination method with the above two classifiers. Results reveal that XGBoost and LightGBM Classifier outperforms the other classifiers in terms of confusion matrix attributes and the proposed hybrid model gives 95.28% accuracy with 13 feature by the lightgbm classifier and 93.4% accuracy by XGBoost classifier with 10 features on the ionosphere dataset.