Table 2.

Time Series algorithms used on the GAN-augmented experimental data set and on the archaeological samples and their results on the raw data set and on the relative data set.

Time Series Classifiers	Configuration of parameters	Combined augmented sample		DS	PTK	FLK Zinj
		Raw data	Relative data
1-KNN (K-nearest neighbour)	metric = 'DTW' (Dynamic Time Warping)	0.86	0.70	P/P (1.00–1.00)	P/P (1.00–1.00)	P/P (1.00–1.00)
2-NN BOSS VS (one nearest neighbour bag-of-SFA-symbols vector space)	word_size = 2–4, n_bins = 4	0.82	0.71	P/P	S/P	P/P
3-SAX-VSM (Symbolic Aggregate ApproXimation and Vector space Model)	word_size = 6, n_bins = 2, strategy = 'normal'	N.C.*	0.63	-/P	-/S	-/S
4-BOSS (Bag Of Symbolic-Fourier-Approximation Symbols)	word_size = 4, window_size = 100 classifiers = KNN, LSVM	0.69	0.61	P/P	S/P	P/P
5-WEASEL (Word ExtrAction for time SEries cLassification)	word_size = 7,4 window_sizes = (5/10, 110/80) Classifier = Logistic Regression (solver = 'liblinear')	0.90	0.75	P/P	P/P	P/P
6-MTF (Markov Transition Field)	Classifiers =  Logistic Regression (solver = 'liblinear') Random Forest image_size = 0.1, n_bins = 5–7	0.79	0.62	P/P (0.63/0.56)	P/P (0.53/0.52)	P/P (0.55/0.51)

Results for experiments show the accuracy in the classification of the testing set. Results for the archaeological assemblages show classification according to experimental set (P, primary access; S, secondary access) and probabilities according to reference data set: first (raw data set), second (relative data set). Only KNN and Markov Fields provide probability of classification for the archaeological assemblages. Classification for archaeological data sets is first for raw data (P or S) and then for relative data (P or S).

*No convergence.