Africa wildlife prediction based on custom convolutional neural network

Abstract

In domains like psychology, computer science, and artificial intelligence, facial expression recognition has major ramifications. This paper suggests classifying input animal photos using convolutional neural network (CNN). The CNN layers kernel size has been modified to (2,2), (3,3), (4,4). The primary objective is to examine how the model responds to changes in the CNN layers kernel size and operation. African Wildlife, a data set of four African species, was utilized for the studies. There are 1508 different themes in this collection, divided into 4 groups with 377 photographs each. Different numbers of test photos and training images were used to determine overall performances. The custom CNN model with a kernel size of (3,3) achieved an accuracy of 57.57% on the dataset. According to the experimental findings, having a kernel that is either too large or too tiny may negatively impact the model and result in undesirable poor accuracy. This study could provide suggestions for predicting animals based on the development of convolutional neural networks.