Anemia has become a critical issue in the public health domain, affecting mostly children and pregnant women. Anemia occurs when the level of hemoglobin is reduced below its normal threshold. In children, anemia can slow down their cognitive development and sometimes lead to death, so it requires critical attention. This study aims to detect anemia in children using pallor palm and fingernail images based on a comparative analysis of a Convolutional Neural Network, k-Nearest Neighbor, Random Forest and Support Vector Machine. The raw images were extracted to obtain the region of interest (ROI) using the Triangle Thresholding algorithm and the entropy grayscale image algorithm. The images were segmented for training, validating and testing the model on a ratio of 70:10:20 respectively. After hyperparameter tuning and model regularization, the Convolutional Neural Network (CNN) achieved the highest accuracy of 97.0% on the fingernails as compared to the pallor palm with an accuracy of 83.1% by the CNN. k-Nearest Neighbor (k-NN) achieves the lowest result of 64.8% on the palm which significantly improves to 84.4% on the fingernail images. The outcome of the study recommends that the use of fingernails should be a diagnostic indicator and a primary focus of anemia detection in children through the use of a non-invasive approach since the conjunctiva of the eye can be exposed to falling objects when examined.