Evaluation Of Image Recognition Models In Machine Learning
Alberto Gambino · International journal of high school research · 2023
Given an image, a computer must be able to classify what the image represents.While this task is relatively simple for humans, it might be challenging for computers.Taking K-Nearest-Neighbor (KNN) as a representative of traditional machine learning methods and Convolutional neural networks (CNN), Back Propagation neural networks (BPNN), and deep Belief network (DBN) as examples of deep learning models, this paper compares and analyzes the traditional machine learning and deep learning image classification algorithms.We expose the accuracy of training models on the MNIST images dataset to find out which is the best-performing model in the showcased task and provide a starting playground that could be extended to many other models.Comparing different image recognition models can help advance the field's state of the art.By identifying the limitations of current models and developing new models that address these limitations, researchers can improve the overall accuracy and efficiency of image recognition systems.The experimental results show that CNN is the best of the four algorithms in image classification accuracy when using the MNIST dataset.What surprised us was not just that a classical machine learning model, KNN, outperformed two deep learning models, BPNN and DBN, but also that DBN performed better than BPNN.