Improved Arabic Alphabet Characters Classification Using Convolutional Neural Networks (CNN)

Table 7

Experimental results on mixing of Hijja and AHCD through different data augmentation techniques.

Mixed dataset		Data augmentation techniques with using Nadam algorithm
		Rotation		Shifting		Rotation and shifting		Flipping		Gaussian noise
Training	Testing	Training accuracy	Testing accuracy	Training accuracy	Testing accuracy	Training accuracy	Testing accuracy	Training accuracy	Testing accuracy	Training accuracy	Testing accuracy
80(%): AHCD; 20(%): Hijja	20(%): AHCD; 10(%): Hijja	99.62	97.42	99.36	98.02	99.38	98.32	99.77	97.08	99.15	96.74
80(%): Hijja; 20(%): AHCD	20(%): Hijja; 10(%): AHCD	97.36	88.47	93.66	89.07	94.91	90.54	95.21	88.21	96.68	88.49
80(%): AHCD; 80(%): Hijja	20(%): AHCD; 20(%): Hijja	97.27	74.53	97.16	75.13	98.13	78.13	98.16	74.22	96.98	74.02