An attempt at implementing the paper https://www.iieta.org/journals/i2m/paper/10.18280/i2m.190603 with a CNN-based classifier and benchmarking it against the single-layer perceptron approach with hand-picked features as inputs.
Results:
Achieved 87% accuracy as opposed to paper's 94% on a test dataset of size 300