9. Scope for Improvement

SCOPE FOR IMPROVEMENT

The training of the neural network was insufficient due to a) the time given b) computational cost

For further improvement into the neural network, the following optimizations could be made:

1. Usage of multiple epochs (iterations) per batch, preferably 100 with the use of early stopping to prevent overfitting
2. More robust dataset by using larger data pool. The data could be collected using different methods, such as open sourcing
3. Deeper neural network – the neural network used for training was much shallower than industrial-use neural networks, once again due to computational cost. For example, the neural network used was 50 layers deep, however a 150 layered neural network could have a higher degree of accuracy, providing the vanishing gradient problem is taken into account.
4. More classes- once again this could be accomplished using a larger data pool. I collected as many classes of vector-borne diseases as possible using data mining and manual collection, however there was insufficient data for some vector carriers, such as the sandfly. Other vectors I did not collect due to low chance of human interaction, such as the sea snail.
5. Releasing the neural network for public use using a mobile application or website. The neural network must have a high degree of accuracy to be useful in society, however following the steps above, the threshold industrial accuracy of >95% can be reached.