AIN 311 MACHINE LEARNING BLOG 4— CNN MODELS

This week, we prepared our first CNN model, which will be the initial experiment for our project. We utilized two different CNN models, and both models successfully exceeded our target accuracy of 70%.

CNN MODEL — 1

CNN Model — 1 Parameters

EPOCH AND ACCURACY FOR CNN MODEL — 1 TRAIN

CNN MODEL — 1 CONFUSION MATRIX

Glass and Metal classes have the most highest accuracy score for CNN Model-1. But paper class has incredible worse an accuracy score.

Confusion Matrix for CNN Model — 1

CNN MODEL — 1 TEST ACCURACY AND LOSS VALUES

Accuracy for Model — 1 is more higher than our goal accuracy. But 2. Model has more higher accuracy and more lower loss value.

CNN MODEL — 2

CNN MODEL — 2 PARAMETERS

EPOCH AND ACCURACY FOR CNN MODEL — 2TRAIN

CNN MODEL — 2 CONFUSION MATRIX

Plastic and Glass classes have the most highest accuracy score for CNN Model — 2.In our first model, there were classes with very low accuracy however, in our second model, such a situation does not exist.

Confusion Matrix for CNN Model — 2

CNN MODEL — 2 TEST ACCURACY AND LOSS VALUES

Accuracy for Model — 2 is more higher than our goal accuracy and Model — 1 accuracy.

When we examined the test results, we can observe that, albeit by a small margin, our second model has a better accuracy compared to the first one. Additionally, the loss value of our second model is significantly lower than that of the first model. Therefore, we are considering using our second model in the project.