Results Summary

Configuration $1$: $1$ CNN layer, $1$ pooling layer, $2$ fully connected layers, learning rate $=0\mathrm{.}01,$ batch size $=64,$ ReLU activation.

Accuracy: $98\mathrm{.}5\%$

Configuration $2$: $2$ CNN layers, $2$ pooling layers, $2$ fully connected layers, learning rate $=0\mathrm{.}005,$ batch size $=32,$ ReLU activation.

Accuracy: $99\mathrm{.}1\%$

Configuration $3$: $1$ CNN layer, $1$ pooling layer, $3$ fully connected layers, learning rate $=0\mathrm{.}01,$ batch size $=64,$ Tanh activation.

Accuracy: $98\mathrm{.}3\%$

Best Configuration

Configuration $2$: $2$ CNN layers, $2$ pooling layers, $2$ fully connected layers, learning rate $=0\mathrm{.}005,$ batch size $=32,$ ReLU activation.

Prediction Accuracy: $99\mathrm{.}1\%$

Insights

Adding more CNN layers and pooling layers generally improves accuracy.

Lower learning rates can lead to better convergence and higher accuracy.

ReLU activation function performs better than Tanh in this context., in this paragraph yeah so just write about how changing these things affects performance, without using numbers or anything, just be likke this combo works best, talk about what each parameter is