<![CDATA[Deep forest has emerged as a lightweight machine-learning approach for image classification, offering lower computational requirements than many deep learning models. However, its classification performance is highly dependent on the selection of the n_estimators parameter, and determining the optimal value remains a challenge. This study aimed to develop an Extended Deep Forest (EDF) algorithm capable of automatically selecting the optimal number of estimators based on training dataset characteristics. The proposed method employs a data-driven interpolation function derived from the empirical relationship between training dataset size and the optimal estimator count. Experiments were conducted using eight binary image datasets, while the performance of EDF and the conventional deep forest was compared on four independent test datasets. The results showed that EDF consistently achieved higher classification performance on datasets requiring more than two estimators while maintaining comparable computational efficiency. Application of the proposed algorithm to thick blood smear malaria diagnosis achieved an accuracy of 94.05% and an F1-score of 94.03%, demonstrating competitive performance while preserving computational efficiency suitable for resource-constrained environments. In conclusion, EDF provides an effective and automated parameter-selection strategy that improves deep forest performance and supports its practical implementation in lightweight image-classification applications, including malaria diagnosis.]]>
Copyrights © 2026
