Fuzzy Ensemble of Deep Learning Models for Image Classification

Ensemble learning and deep learning are highly effective approaches that often outperform the machine learning models. In ensemble learning, multiple base learning models are combined to create a single model that is more accurate and efficient than the base learning models used in the ensemble. Deep learning makes use of advanced neural network architectures for extracting complex features and patterns to achieve high predictive performance in various domains of computer vision. However, deep learning approaches have inherent limitations such as the need for huge training data, sensitivity to initial conditions, and tendency to overfit with small datasets. These limitations can be addressed by the fusion of ensemble learning and deep learning, a hybrid approach known as deep ensemble learning. In this study, we have proposed a stacking-based fuzzy ensemble technique that combines several deep learning models with the help of fuzzy neural network-based meta-learning algorithm. The evaluation process is conducted on the flower image dataset from the Kaggle data repository, and according to the results, the proposed fuzzy ensemble technique surpasses state-of-the-art deep learning models and ensemble methods like majority voting, averaging, and weighted averaging. The proposed technique has attained the maximum accuracy of 92.61% for 5-class data, which demonstrates its effectiveness.