As part of my coursework, I developed a binary classification model that differentiates between malign and benign melanoma. To put this into practical use, I converted into a tflite model to be deployed onto a raspberry pi for practicality.
The dataset used is the Kaggle Skin Cancer Dataset
To interpret this confusion matrix, benign is set to 0 while malign is set to 1 (as one can see from the notebook)
With this key, we can decipher that it has 41% chance of detecting both benign and malign melanoma in correspondence with the ground truth.
A 3.6% chance of detecting a malign melonoma as benign and 13.18% chance of detecting a benign melonoma as malign.
With respect to medical standards, having a low FN score is very essential as the device shouldn't fail to acknowledge fatality.
To support the development of the product, a Fusion 360 Sketch was also made to visualise the placements of the different components such as
battery pack,
Raspberry Pi,
and the camera housing
To view the sketch, visit here
