This project focuses on achieving exceptional classification accuracy through feature selection and Auto-ML techniques. We explore various feature selection methods and employ Auto-ML with multiple estimators, presenting a comparative analysis of their performance.
Key Highlights:
-
Feature Selection Diversity: We meticulously selected features using six correlation methods, including ANOVA, Kendall, Mutual Information (MI), Spearman, and Point. These methods provide a comprehensive set of features for classification.
-
Feature Combination Exploration: We analyzed a whopping 2,577 feature combinations using recursive and sequential techniques, along with the application of over 175 metaheuristic optimizers. This exhaustive exploration enhances our understanding of the feature space.
-
Auto-ML Deployment: We leveraged Auto-Machine Learning (Auto-ML) techniques with eight different estimators and achieved a remarkable accuracy of 99.87% using the Random Forest multiclassifier.
-
Comparative Insights: In our study, we discovered that while Auto-ML methods offer faster results, a slight edge in accuracy (1-1.75%) can be obtained using Long Short-Term Memory (LSTM) models, underscoring the trade-offs between speed and precision in automated machine learning.
Contribution: This project provides a valuable resource for researchers and practitioners in the field of machine learning and feature selection. It offers a wealth of feature selection methods, extensive feature combination analysis, and insights into the performance of Auto-ML compared to traditional LSTM models.