5. Underfitting and Overfitting

One problem in my neural network that I immediately identified before even beginning to program was underfitting. Underfitting is a problem associated with the neural network failing to capture the complexity of the data resulting in poor training testing and poor unseen data testing. With little data in the majority of the insect genera, I found it difficult to keep bias at a minimum and have a high variance in the dataset. So, I decided to increase my dataset through image augmentations. Although this solution is not ideal, as it is essentially variations of the already given images, it was the best solution with the dataset I mined, as it would be near impossible to create a larger dataset of images due to the specific nature of data required. Attempts were made at finding datasets of vector carriers, however, none seemed to be available, or even had been created before. I hope to improve this flaw in the system by using the data collected from the use of the neural network to train the neural network, thus improving its accuracy.

The opposite of underfitting is overfitting, occurring due to overtraining and poor optimization. Fortunately this did not pose any difficulty during the training of the neural network.