Mixed Script Identification Using Automated DNN Hyperparameter Optimization

Table 5

Accuracy comparison with parameter optimization though Parzen estimator.

Glove embedding with Parzen estimator
Dataset	Classification algorithms
	LSTM	Bi-LSTM	GRU	Bi-GRU
Eng-Roman Urdu	0.552	0.843	0.792	0.851
Saraiki-Hindi	0.456	0.791	0.81	0.823
Bengali-Hindi	0.376	0.754	0.812	0.851
Eng-Bengali	0.552	0.758	0.792	0.889
Saraiki-English	0.612	0.802	0.797	0.841
Saraiki-Roman Urdu	0.756	0.812	0.779	0.862
Saraiki-Bengali	0.521	0.723	0.782	0.834
Eng-Bengali-Saraiki-Hindi-Roman Urdu	0.379	0.669	0.76	0.802

Bold indicates the maximum performance of the algorithm.