Updated: May 3
It has been a slow progress but so far this is what we have been able to achieve
Create an algorithm for basic component detection.
Improve accuracy to 98% on train/validation set
The next step is to test on an unseen data set. Additionally the previous tests revealed the need for an unclassified category as well since there were instances in data set that did not fall in any of the main categories an they were mislabeled by the system causing an increased in error rate.