Table 1
Summary of results of both automatic and manual protocol searches on the six online portals. Note that the “Sum of unique totals” refers only to the sum of the number of non-duplicate results from each search; however, there were many publications present in both searches (see below), which would further reduce the sum total of unique items.
| BIBLIOGRAPHICAL DATABASE | KEYWORDS MATCHES |
|---|---|
| Automatic screening: | |
| Web of Science | 969 |
| PubMed | 413 |
| Tübingen University Library | 51 |
| German Archaeological Institute | 24 |
| German National Library | 3 |
| Total unique | 558 |
| Manual screening: | |
| Google Scholar | 300 |
| Total unique | 285 |
| Total | 1760 |
| Sum of unique totals | 730 |
Figure 1
Review process from source selection to analysis. Inspired by the PRISMA 2020 flow diagram (Page et al. 2021). Reason 1 = Ineligible with automation tool; Reason 2 = Non-English record; Reason 3 = Full text not accessible; Reason 4 = Non-journal-based publications; Reason 5 = Absence of abstract; Reason 6 = Archaeology and machine learning keywords from the list not present in the text; Reason 7 = Archaeology and machine learning keywords from the list are not present in the abstract or in the title; Reason 8 = Preliminary exclusion (i.e. no access to publication, publications or contribution by current authors, entire books, non-academic reports, preprints, reviews or theoretical papers, potentially predatory journal); Reason 9 = Excluded based on the title; Reason 10 = Excluded based on abstract; Reason 11 = Excluded based on the full text first reading; Reason 12 = Full text does not involve archaeological research; Reason = 13 Full text does not involve machine learning methods as defined in our protocol; Reason 14 = Conflicts of interest (publication by the authors of this review or in which the authors contributed); Reason 15 = Theory or review paper. Figure created using Microsoft Word and Inkscape.
Table 2
The nine features collected systematically from the review.
| FEATURE | NUMBER OF CATEGORIES |
|---|---|
| Model | 70 |
| Best model | 17 |
| Family | 9 |
| Subfield | 15 |
| Input data | 11 |
| Evaluation | 3 |
| Task | 19 |
| Result | 5 |
| Pre-training | 4 |
Figure 2
The fourth field of information recorded in the review presents significant characteristics to explain variation in machine learning applications in archaeology and their related classes/categories. One study case might have been attributed to several subfields or architecture categories. Figure generated with R 4.2.2 (code available in supplementary material 3) and additional editing with Inkscape.
Figure 3
Number of publications per year between 1997 and 2022, in light blue the articles published after 2018 concentrated more than 80% of the publications. The dashed line represents publications from 1 January 2023 to 31 September 2024. Figure generated with R 4.2.2 (code available in supplementary material 3) with additional editing in Adobe Illustrator.
Table 3
The ten most represented journals and their h-index and Impact factor (IF) score and total score by the number of articles, n = 135. Metrics were consulted on 14/07/2024 on the paper website for the impact factor or on SJR for the h-index (supplementary file).
| JOURNAL | NUM. OF ARTICLES | H-INDEX | N ⋅ H-INDEX | IF | N ⋅ IF |
|---|---|---|---|---|---|
| Remote Sensing | 15 | 193 | 2895 | 4.2 | 63 |
| Journal of Archaeological Science | 14 | 152 | 2128 | 2.6 | 36.4 |
| PLOS One | 11 | 435 | 4785 | 3.75 | 41.25 |
| Scientific Reports | 6 | 315 | 1890 | 3.8 | 22.8 |
| Journal of Computer Applications in Archaeology | 5 | 15 | 75 | N/A | N/A |
| Archaeological Prospections | 4 | 46 | 184 | 2.1 | 8.4 |
| Journal on Computing and Cultural Heritage | 3 | 35 | 105 | 2.7 | 8.1 |
| Archaeological and Anthropological Sciences | 3 | 42 | 126 | 2.14 | 6.42 |
| Palaeogeography Palaeoclimatology Palaeoecology | 3 | 177 | 531 | 2.6 | 7.8 |
| Virtual Archaeology Review | 3 | 17 | 51 | 1.6 | 4.8 |
Figure 4
Number of articles published per country based on the country of the first author’s affiliation. Figure generated with R 4.2.2 (code available in supplementary material 3).
Figure 5
(A) Number of articles from each archaeological subfield between 1997 and 2022. (B) Number of articles from each architecture class between 1997 and 2022. Empty bar charts represent the number 1. Figure generated with R 4.2.2 (code available in supplementary material 3).
Figure 6
Tree map of the different models seen in our corpus as well as the family of models they belonged to in our categorisation. Figure generated with R 4.2.2 (code available in supplementary material 3).
Figure 7
(A) The five more represent classes of input data among the reviewed papers, n = 148. (B) Results of the reviewed papers according to the authors or presented results, n = 147. Figure generated with R 4.2.2 (code available in supplementary material 3).
Figure 8
Alluvial diagram of the different tasks in the analysed studies on the left, the related architecture of machine learning models on the right and the evaluation process in the background. Tasks and architectures poorly represented (n < 5) have been classified as “others”. A study might have applied numerous models, or its research objectives could be classified into more than one task. In such cases, we created multiple entries for each paper where applicable (see supplementary material). Figure generated with R 4.2.2 (code available in supplementary material 3).
Figure 9
Alluvial diagram of the different tasks in the analysed studies on the left, the related archaeological subfields on the right with the evaluation process in the background. Tasks and subfields poorly represented (n < 5) have been classified as “others”. A study might have been attributed to several subfields, or its research objectives could be classified into more than one task. In such cases, we created multiple entries for each paper where applicable (see supplementary material). Figure generated with R 4.2.2 (code available in supplementary material 3).
Figure 10
Alluvial diagram of the different tasks in the analysed studies on the left, the related results on the right with the evaluation process in the background. Tasks poorly represented (n < 5) have been classified as “others”. A study might have its research objectives classified into more than one task. In such cases, we created multiple entries for each paper where applicable (see supplementary material). Figure generated with R 4.2.2 (code available in supplementary material 3).
Figure 11
Workflow adapted to machine learning solutions applied to archaeological problematics. Figure generated with Microsoft Word and Inkscape.
Annexe 1
List of algorithms used present in the papers under review organized study cases reviewed, organised by the approach and family of analysis methods they were categorised in, along with their abbreviations and number of use. In the case the model was compared to others, we highlighted the, number of times he performed used in our corpus, and number of times selected by the authors of a study case as the best algorithm when performing a comparison of various models.
| FAMILY | MODEL | ACRONYM | NB. OF USES | NB. OF TIME BEST | |
|---|---|---|---|---|---|
| Artificial Neural Network | Feedforward Neural Network | FNN | 23 | 4 | |
| Convolutional Neural Network | CNN | 14 | 1 | ||
| Residual Neural Network | ResNet | 12 | 2 | ||
| Mask Region-based Convolutional Neural Network | MR-CNN | 9 | 1 | ||
| Faster Region-based Convolutional Neural Network | FR-CNN | 8 | 0 | ||
| Visual Geometry Group | VGG | 8 | 2 | ||
| U-Net | U-Net | 7 | 4 | ||
| Inception Network | INC | 4 | 1 | ||
| AlexNet | AlexNet | 3 | 0 | ||
| RetinaNet | RN | 3 | 0 | ||
| YOLO | YOLO | 3 | 0 | ||
| DeepLabv3+ | DL3 | 2 | 0 | ||
| Semantic Segmentation Model | SegNet | 2 | 0 | ||
| Adaptive Deep Learning for Fine-grained Image Retrieval | ADLFIG | 1 | 0 | ||
| Bidirectional Encoder Representations from Transformer | BERT | 1 | 0 | ||
| Bidirectional Gated Recurrent Unit | BiGRU | 1 | 0 | ||
| Bidirectional Long Short-Term Memory Network | BiLSTM | 1 | 0 | ||
| Dynamic Graph Convolutional Neural Network | DGCNN | 1 | 0 | ||
| DenseNet201 | DN201 | 1 | 0 | ||
| Generative Adversarial Network | GAN | 1 | 0 | ||
| Jason 2 | JAS2 | 1 | 0 | ||
| Neural Support Vector Machine | NSVM | 1 | 0 | ||
| Region-based Convolutional Neural Network | R-CNN | 1 | 0 | ||
| Simple Network | SimpleNet | 1 | 0 | ||
| Single Shot MultiBox Detector | SSD | 1 | 0 | ||
| Bayesian Classifier | Naïve Bayes | NB | 11 | 0 | |
| Maximum Entropy | MaxEnt | 2 | 0 | ||
| Decision Trees and Rule Induction | C5.0 | C5.0 | 7 | 2 | |
| C4.5 | C4.5 | 4 | 0 | ||
| Decision Tree/Classification Tree | DT | 4 | 0 | ||
| Conditional Inference Trees | CTREE | 2 | 0 | ||
| Iterative Dichotomiser 3 | ID3 | 2 | 0 | ||
| Classification And Regression Tree | CART | 1 | 0 | ||
| Fast and Frugal Tree | FFT | 1 | 0 | ||
| Learning with Galois Lattice | LEGAL | 1 | 0 | ||
| Representative Trees | REPTree | 1 | 0 | ||
| Random Trees | RT | 1 | 0 | ||
| Ensemble Learning | Random Forest | RF | 54 | 20 | |
| Adaptative Boost | AdaBoost | 2 | 0 | ||
| Stochastic Gradient Boosting | SGB | 2 | 1 | ||
| eXtreme Gradient Boosting | XGB | 2 | 1 | ||
| Bootstrap Agreggating | BAgg | 1 | 0 | ||
| Discrete Super Learner | dSL | 1 | 0 | ||
| Fast Random Forest | FRF | 1 | 0 | ||
| Gradient boosting Regression Tree | GboostRT | 1 | 0 | ||
| LogitBoost | LB | 1 | 0 | ||
| Quantile Random Forest | QRF | 1 | 0 | ||
| Sequential Backward Selection-Random Forest Regression | SBS-RFR | 1 | 1 | ||
| Synthetic Minority Over-sampling Technique Boost | SMOTEBoost | 1 | 0 | ||
| Synthetic Minority Oversampling Technique + Edited Nearest Neighbor Rule | SMOTEENN | 1 | 0 | ||
| Super Learner | SP | 1 | 1 | ||
| Viola-Jones Cascade Classifier | VL-CC | 1 | 0 | ||
| Genetic Algorithm | Genetic Algorithm | GA | 1 | 0 | |
| Linear Classifier | Support Vector Machine | SVM | 26 | 2 | |
| Structured Support Vector Machine | SSVM | 1 | 0 | ||
| Nearest Neighbour Classifier | k-nearest neighbors | kNN | 19 | 1 | |
| Weighted k-nearest neighbors | kkNN | 3 | 0 | ||
| Polynomial Classifier | Support Vector Machine with Radial Basis Function Kernel | SVMr | 7 | 1 | |
| Unsupervised Learning and Clustering | Affinity Propagation | AF | 1 | 0 | |
| Hierarchical Cluster-Based Peak Alignment | CluPA | 1 | 0 | ||
| Databionic Swarm | DBS | 1 | 0 | ||
| Expectation-Maximisation Clustering | EMC | 1 | 0 | ||
| Graph-based Semi-Supervised Learning | GSSL | 1 | 1 | ||
| Iterative Closest Point | ICP | 1 | 0 | ||
| Iterative Self-Organizing Data Analysis | ISODATA | 1 | 0 | ||
| Nearest Centroid | NC | 1 | 0 | ||
| Simple Linear Iterative Clustering | SLIC | 1 | 0 | ||
| Self-Organizing Map | SOM | 1 | 0 | ||
| Tilburg Memory-Based Learning | TiMBL | 1 | 0 | ||
| Time series clustering | TSC | 1 | 0 |