Applying machine learning tools for automated behaviour classification in invasive lionfish and comparison with human observations

Applying machine learning tools for automated behaviour classification in invasive lionfish and comparison with human observations | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Applying machine learning tools for automated behaviour classification in invasive lionfish and comparison with human observations Amirreza Khodaparast Kelidbari, Katelyn Moffat, Iris George, Nicola Smith This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7153188/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 16 Feb, 2026 Read the published version in Marine Biology → Version 1 posted 5 You are reading this latest preprint version Abstract Modern neuroscience and ecology are increasingly adopting machine learning (ML) methods to automate the tracking and classification of animal behaviour. These techniques are particularly valuable for quantifying fitness‑related behaviours such as hunting in invasive predators. Here, we evaluate the effectiveness of two ML tools, DeepLabCut for pose estimation and SimBA for random‑forest behaviour classification, at distinguishing four behaviours in invasive lionfish ( Pterois volitans and P. miles ): hovering, resting, swimming, and hunting, and benchmark the ML outputs against annotations from trained human observers. We also introduce a customized, user‑friendly feature‑extraction script tailored to lionfish. The script converts positional landmark coordinates extracted by DeepLabCut into a comprehensive set of kinematic metrics (e.g., body‑angle variance, fin‑beat frequency), essential for behaviour classification, as SimBA relies on these metrics rather than raw body‑part positions. A companion GitHub guide further clarifies which specific metrics are most informative under different behavioural scenarios. To our knowledge, this is the first study to explore how a wide‑angle camera lens influences the DeepLabCut–SimBA workflow. Behavioural trials, conducted in controlled aquarium settings, showed that the models classified high‑motion behaviours (hunting and swimming) with high precision and recall, likely owing to distinctive kinematic signatures. In contrast, low‑motion behaviours such as hovering and resting were harder to detect because of subtle movement cues, occasional suboptimal body orientation, and distortions introduced by the wide‑angle lens. While applicable to other mid-bodied fishes, lionfish were chosen due to their significant ecological impact, where quantifying behaviours like hunting can aid invasive species management and reef conservation. Lionfish Behaviour classification Pose estimation Machine learning DeepLabCut SimBA Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Introduction Invasive lionfish and its ecological impacts A particularly successful pair of marine invaders are the Indo-Pacific lionfishes ( Pterois volitans and P. miles ), which have colonized the western Atlantic Ocean. Introduced via the aquarium trade in the mid-1980s, lionfish were first documented along Florida’s east coast in the early 1990s. By the early 2000s, they had rapidly established populations along the U.S. eastern seaboard and subsequently spread throughout the Caribbean and Gulf of Mexico. Lionfish are now established as far north as North Carolina in the US and as far south as Brazil (Morris & Akins, 2009 ; Soares et al. 2023 ). They are also currently invading the eastern Mediterranean (Bernardi et al. 2024 ). Able to thrive in diverse habitats, from estuaries and shallow coral reefs to mesophotic reefs at depths exceeding 200 meters, lionfish have demonstrated remarkable adaptability, making their invasion one of the most rapid and geographically extensive marine fish invasions documented to date (Green et al. 2012 ; Albins & Hixon, 2008 ). Lionfish feed on up to 70 species of native reef fishes and numerous crustaceans in their invaded range, many of high ecological and/or commercial value (Côté and Smith, 2018 ). They use an ambush strategy characterized by slow stalking, sudden strikes, and flared pectoral fins to corner prey (Green et al. 2019 ). This technique has been highly effective in the Atlantic, where naïve prey have not evolved defenses against this novel predator. In some regions, such as The Bahamas, lionfish predation has led to 65% declines in native prey fish biomass on natural reefs within two years of establishment (Green et al. 2012 ). Fish reductions have, in some cases, resulted in localized extirpations of vulnerable prey species and the replacement of native mesopredator biomass on some reefs (Côté & Smith, 2018 ). Notably, field surveys in The Bahamas have observed lionfish occupying the ecological niche once held by native groupers, with uncertain consequences for reef communities (Côté & Smith, 2018 ). Given the substantial, negative ecological impacts of lionfish, accurately quantifying their predatory behaviours is essential for effective invasive species management and coral reef conservation. Machine learning in behavioural analysis Modern neuroscience and ecology are increasingly leveraging machine learning (ML) techniques for automated animal behaviour tracking and classification (Christin et al. 2019 ). Traditionally, researchers depended on manual observation and scoring of behaviours, an approach that is labour-intensive, requires extensive training, and can be prone to observer bias (Mathis et al. 2018 ). By contrast, ML-based approaches offer a high-throughput alternative, enabling large volumes of video data to be processed while maintaining consistent scoring standards at reduced costs (Nilsson et al. 2020 ). Two popular open-source ML tools for automated behavioural analysis are DeepLabCut and SimBA. DeepLabCut uses deep neural networks to estimate animal poses in a markerless fashion, enabling precise tracking of body parts (Mathis et al. 2018 ). SimBA (Simple Behavioural Analysis) expands on such pose data with a user-friendly platform for training supervised classifiers to detect specific behaviours using random forest models (Nilsson et al. 2020 ). Random forests, which are ensembles of decision trees trained on random subsets of features and data, effectively mitigate overfitting that single decision trees often face (Zhou, 2021). Both tools have robust user communities that offer support for setup, troubleshooting, and customization. Although DeepLabCut and SimBA have been used and validated in many studies involving rodents (Weber et al. 2022 ; Chanthongdee., 2024), relatively few have tested these tools on medium-bodied fishes, including invasive lionfish (Fig. 1 ). Study objectives In this study, we conduct a controlled experiment using an aquarium setup to evaluate the effectiveness of a machine learning workflow for detecting and quantifying behaviours in mid-bodied fishes, specifically invasive lionfish. Our approach combines deep neural networks for pose estimation (via DeepLabCut) with random forest classifiers (using SimBA) to automatically classify four behaviours: hovering, resting, swimming, and hunting. We aim to address two questions. First, can machine learning methods accurately quantify lionfish behaviours using a wide-angle lens? Second, how does the performance of our automated classification compare to human observations? Methods Experimental setup and lionfish behavioural trials Behavioural trials occurred twice a day, at dawn and midday, for two days during July–September 2024 at the Cape Eleuthera Institute on Eleuthera Island, The Bahamas. Each trial consisted of a 75 G aquarium tank that was separated into two sections by opaque and transparent dividers (Fig. 2 ). An adult lionfish (mean ± se = 20.38 cm ± 1.43 total length, TL) was placed in one part of the tank while three to four juvenile, native prey fishes (mean ± se = 6.24 cm ± 0.21 TL) (princess and striped parrotfish, Scarus taeniopterus and S. iseri ) were placed in the other part. Fish shelters in the form of rocks and PVC pipe were also placed in each section of the tank (Fig. 2 ). At the start of a trial, the opaque divider was removed while the transparent divider remained in place. This setup allowed the fishes to see one another but prevented physical contact. Once a trial began, 30 minutes of video footage of lionfish behaviours was recorded using GoPro Hero11 cameras in wide-angle mode, set to 4K resolution at 30 frames per second. Footage was then downscaled to 1080p with FFmpeg for computational efficiency. All video files were labelled and stored, totaling approximately 12 hours of footage of 7 adult lionfish. Lionfish were never reused in the experiment except between dawn and mid-day trials. All behavioural trials were conducted under an animal care permit (#30019689) from Concordia University, Quebec, Canada. Human observations Two human observers were trained to classify a pre-defined set of lionfish behaviours and record lionfish position relative to the dividers (i.e., near, mid, or far) at 30-second intervals for each 30-minute trial, across 24 total trials, resulting in 1,219 data points for comparison with machine generated predictions. Observers officially recorded their data once there was at least 95% agreement between them (inter-observer reliability) and a repeatability score of 95% within each observer (intra-observer reliability). To confirm repeatability, each observer rewatched and rescored a portion of lionfish footage until consistency in behaviour scoring was achieved across multiple assessments. The predator (invasive lionfish) was separated from the prey (native parrotfish) by both transparent dividers (screen mesh) and an opaque divider (black sponge). At the start of the experiment, the opaque divider was removed so that the fishes could see but not physically contact each other, and lionfish behaviors were recorded. An ultra wide camera captured the entire tank setup. The black grid in the background served as a spatial reference for estimating lionfish position and movement. DeepLabCut pose estimation DeepLabCut (Mathis et al. 2018 ) was used to track lionfish movements (Fig. 3 ). A total of 1,000 video frames were extracted from pilot recordings captured in wide-angle mode, which introduced greater variation in lionfish size, shape, and orientation across the frame. While ~ 300 frames are typically sufficient for training, the increased visual variability caused by the wide-angle perspective required a larger set of annotated frames to ensure the model could accurately generalize across diverse poses and viewing angles. Frame annotation was performed in Napari to label specific anatomical points on each frame (Fig. 3 ). A ResNet-101 backbone was then trained on a cloud-based NVIDIA L4 GPU for four hours, using DeepLabCut’s default training parameters (including max_input_size: 1500, min_input_size: 64, global_scale: 0.8, pos_dist_thresh: 17, pafwidth: 20, scale_jitter_lo: 0.5, scale_jitter_up: 1.25) and adjusting only the batch size to accommodate GPU memory constraints. The performance of the pose estimation model was evaluated following training with DeepLabCut for 30,000 iterations, at which point the test error had stabilized. Of the annotated frames, 95% were allocated to the training set, while the remaining 5% were used for testing (shuffle 1). With images recorded at 1 920 × 1 080 px, the network achieved an average training error of 6.21 px and a test error of 5.32 px (p-cutoff = 0.65). The < 1-px train–test gap meets the DeepLabCut guideline that comparable errors indicate good generalisation (Mathis et al. 2018 ). In absolute terms, 5.32 px corresponds to 0.28% of frame width — within the 2–6-px range reported for benchmark rodent and fish datasets (Mathis et al. 2018 ; Pereira et al. 2019 ). These results showed that the model was able to accurately predict body part locations when the fish has the optimal orientation (side view) towards the camera. The trained network was then used to analyze additional lionfish videos, again employing the NVIDIA L4 GPU to ensure efficient inference. Lionfish body parts with distinctive shapes and colours, such as the head and certain fin tips, showed high detectability in DeepLabCut. In contrast, some spines and fins had lower detectability due to frequent overlapping and the lack of uniquely identifiable features. The dorsal spines, however, showed relatively higher detection likelihood and may be more accurately tracked using an optimized setup. The mid-body region was often covered by the pectoral fins, resulting in consistently low detection likelihood. SimBA workflow CSV pose data extracted from DeepLabCut were imported into the SimBA platform for subsequent behavioural analysis. All pose-estimation outputs were first preprocessed to remove outliers by applying a movement criterion of 1.25 body‐lengths per frame (distance between body parts 1 [mouth] and 10 [caudal fin tip]). This threshold was chosen based on pilot analyses in which we systematically varied the movement criterion from 0.7 BL (body length)/ frame (SimBA’s default movement criterion) to 1.5 BL/frame. Thresholds 1.5 BL/frame passed most false jumps. At 1.25 BL/frame we achieved the optimal balance to remove the outliers while maintaining genuine rapid movements. Positional trajectories were then smoothed using a Savitzky–Golay kernel (Savitzky, A. et al. 1964). Based on each frame’s body-part coordinates, along with the known frame rate and pixels-per-millimetre ratio, SimBA’s feature extraction step generated a set of kinematic and morphological metrics. A modified version of the default zebrafish extraction file available on the SimBA website was used (Table S1; https://doi.org/10.5281/zenodo.15734546 ) to accommodate additional lionfish-specific key points. Selecting features via permutation importance before retraining a Random Forest significantly improved short-term load-forecasting performance compared to using all variables (Huang, N. et al. 2016 ). To evaluate the utility of additional features, we manually added them to the feature extraction file and assessed their contribution using the feature importance rankings in SimBA. Features that failed to improve behavioral differentiation were removed, while those that improved class separation were retained, having ranked within the top 200 most informative features out of 893 in at least one behaviour extracted, based on permutation importance scores. The platform computed features such as rotation/orientation features, path metrics, pairwise distances, measures of velocity and acceleration, and more, resulting in 893 extracted features. SimBA’s integrated ROI definition tool was then employed to partition the experimental tank into spatial zones (Fig. 4 ), and these zone assignments were appended to the extracted features to link behaviour with location. A two-hours subset of pilot video footage was manually annotated in SimBA (Table 1 ) to label the presence or absence of four defined behaviours. Kinematic features extracted from pose data were used as input for a supervised machine learning classifier. A random forest model was trained using the following parameters: 2,000 estimators, max_features = sqrt, criterion = entropy, test_size = 0.3, and train-test split type = FRAMES. The min_samples_leaf was set to 1, with no minimum or maximum under sampling. Balanced class-weight adjustments were used to account for class imbalances in the presence and absence of resting, hunting, hovering, and swimming behaviours. Classification thresholds were manually determined using the precision-recall curve and the interactive probability vs. frame number plot to find the best balance between precision and recall. The trained model was subsequently applied to additional lionfish csv files to generate behaviour classifications. Fish position was estimated using the weighted average of selected body parts, with weights based on their detection probabilities. The experimental tank was divided into five zones based on vertical and horizontal banding, providing structured regions to analyze lionfish movement and positioning. The "bottom" zone spans the entire lower row across the tank, with "sub-bottom" defined as half of that band. The "near" zone includes the first two vertical bands adjacent to the transparent divider, while the "mid" zone extends from vertical band two through band five. The top zone covers any area above the first horizontal band. The "far" zone covers vertical bands six through eight. Table 1. Behavioural classifiers and operational definitions for lionfish. Definitions of four distinct lionfish behaviours: hovering, resting, swimming, and hunting. To assess the performance of the behavioural classifiers, we computed three standard evaluation metrics in the absence and presence of the behaviours: recall, precision, and F1 score. a. Recall (absence) - measures how well the classifier correctly identifies when the behaviour is truly absent. $$\:Recall\left(absence\right)\:=\frac{True\:Negatives}{True\:Negatives\:+\:False\:Positives}$$ b. Recall (presence) - measures how well the classifier detected the behaviour when it was truly present. $$\:Recall\left(presence\right)=\frac{True\:Positives}{True\:Positives\:+\:False\:Negatives}\:$$ 2a. Precision (absent) - indicates how often a prediction of absence was correct. $$\:\:Precision\left(absent\right)=\:\frac{True\:Negatives}{True\:Negative\:+False\:Negative}$$ 2b. Precision (present) - indicates how often the classifier was correct when it predicted that a behaviour was present. $$\:Precision\left(present\right)\:=\frac{True\:Positive}{True\:Positive\:+\:False\:Positive}\:$$ 3. F1 score - provides a single metric that balances precision and recall. $$\:F1=\:\frac{2\:\times\:\:Precision\:\times\:\:Recall}{Precision\:+\:Recall}$$ Results Human agreement and reference set validation Two human observers independently annotated the four behaviours: hovering, resting, hunting, and swimming, with both inter-observer agreement and intra-observer repeatability ≥ 95%. For spatial classification (near or far zones), agreement reached 99% for both inter- and intra-observer comparisons. External validation: classifiers vs. human observations consensus External validation measures how well the trained classifier reproduces labels made by independent human observers on truly unseen data and setups. For each behaviour we built a confusion matrix by tallying true negatives (TN), false negatives (FN), true positives (TP) and false positives (FP; Table S2). Using the confusion matrices (Table S1), precision, recall and F1 were separately calculated for the presence and absence of that behaviour. Classifier performance differed across behaviours (Table 2). Specifically for hunting, the model correctly identified 694 of 722 true events (Table S2), yielding 86.2% precision and 96.1% recall (F1 = 0.91) (Table 2a). The negative class was also reliable, with 92.9% precision and 77.7% recall, indicating relatively few false alarms (Table 2a). Swimming showed a similar profile: recall remained high (85.3%), capturing most swim bouts, but precision was moderate (64.1%) and showed some mislabelling; the resulting F1score was 73% (Table 2b). In contrast, rest and hover were not as reliable. Rest was detected in only 34.8% of true instances and with 33.0% precision (F1 = 0.34) (Table 2d), while hover achieved 30.6% recall and 57.8% precision (F1 = 0.40) (Table 2c). In both cases, precision and recall for the absence class exceeded 84% (Fig. 2 c − 2d). For distance from the prey estimation, when the fish was near the prey, the location was rarely misclassified with 95.0% precision, 96.0% recall (Table 2e). When the fish was in the mid or far zones (Fig. 4 ), its location was detected only two‑thirds of the time, with 82.0% precision, 66.0% recall, and F1 = 0.73 (Table 2e). Internal cross‑validation: model generalisation Within SimBA, we withheld 30% of our author‑labelled frames, ensuring whole behavioural bouts remained intact to prevent temporal leakage, and trained on the remaining 70%. SimBA then automatically generated per‑behaviour confusion matrices (TN, FP, FN, TP) on this internal test split and calculated precision, recall and F1 for both presence and absence of each behaviour. Hunting (Table 3a) was detected with high fidelity: presence precision reached 97.7% and recall 90.8% (F1 = 94.1%), while the absence class also performed strongly (82.5% precision, 95.4% recall; F1 = 88.5%), indicating very few false positives. Swimming (Table 3b) exhibited a similarly robust absence profile—98.1% precision and 96.9% recall (F₁ = 97.5%) and a good presence performance (73.3% precision, 81.8% recall; F1 = 77.3%), with some misclassifications of actual swim bouts. In contrast, resting and hovering were far less reliable. Rest events were flagged with just 49.4% precision and 25.2% recall (F1 = 33.4%) (Table 3d), while hover classifier achieved only 21.8% precision and 56.3% recall (F1 = 31.4%) (Table 2c). Despite these low presence scores, absence metrics for both behaviours remained high (hover: 98.1% precision, 91.9% recall; rest: 82.5% precision, 93.1% recall). The close agreement between F1 values of internal and external validations indicates that our pilot training videos effectively captured nearly the full diversity of behavioural poses encountered in independent datasets. Threshold analysis We also looked at classifier performance by the precision-recall curve across behaviours. A precision–recall curve plots how precision and recall trade off as the threshold is changed in a classifier. The classifier for hunting behaviour showed excellent performance across thresholds, with both precision and recall remaining consistently high (Fig. 5 a). The F1 score peaked around ~ 0.5, with a strong balance between detecting true hunting events and minimizing false positives (Fig. 5 a). In contrast, the swim classifier showed a clear precision-recall trade off, with precision increasing and recall decreasing as the threshold rose (Fig. 5 b). The F1 score peaked around a threshold of 0.55 at a ~ 0.75 performance, which indicates good overall performance (Fig. 5 b). The resting behaviour classifier exhibited a steep precision–recall trade off driven by overlapping kinematic features with non‑rest frames (Fig. 5 d). At a discrimination threshold of 0.00, the model achieved maximal recall (100%) but low precision (≈ 40%), yielding an F1 of 0.57 (Fig. 5 d). Increasing the threshold to 0.10 boosted precision to 66% while maintaining recall at 90%, producing the peak F1 of 0.79 (Fig. 5 d). Beyond a threshold of 0.20, recall fell below 30% and precision plateaued around 50%, causing F1 to decline to ~ 0.34 (Fig. 5 d). Lastly, the hover classifier showed a very shallow trade off: at a threshold of 0.00 it achieved 100% recall but only ~ 8% precision (F1 ≈ 0.15), and at the optimal threshold (~ 0.35) precision rose to ~ 22% while recall fell to ~ 68%, yielding a peak F1 of just ~ 0.33 (Fig. 5 c). Beyond the threshold of 0.40, recall continued to decline toward zero even as precision plateaued around 20%, driving F1 back down (Fig. 5 c). Behaviour composition and misclassification patterns Behaviour distribution analysis revealed significant class imbalance, with hunting being the most frequent behaviour observed (Fig. 6 a). Spatial analysis revealed that resting occurred mostly at tank edges, areas prone to increased lens distortion (Fig. 6 b) while hovering is commonly misclassified as hunting (Fig. 6 c). Discussion We combined DeepLabCut’s deep learning–based pose estimation with SimBA’s random forest classifiers to detect four lionfish behaviours. The models accurately classified high-motion behaviours (hunting and swimming) achieving high precision and recall. In contrast, low-motion behaviours (hovering and resting) were harder to detect. We also found that while tracking remained reliable near the frame center, distortion at the edges reduced pose accuracy and degraded classification performance. Discrete vs continuously variable behaviour Because SimBA assigns each video frame a continuous likelihood score for behavioural categories rather than strictly discrete labels, this workflow is suitable for studying transitional behaviours. Researchers can quantitatively track how an animal’s likelihood of exhibiting a certain behaviour gradually increases or fades away, capturing the temporal dynamics inherent in behavioural transitions. For instance, by plotting the likelihood curve through time, one can visualize a fish's hunting likelihood rising incrementally during preparatory motions, declining briefly if the fish hesitates or initiates an alternative action, and then increasing again upon executing a full predatory lunge. This continuous probabilistic output allows investigators to examine transitional states in detail, quantify their similarity to established behaviour categories, and explore ecological hypotheses related to behavioural plasticity and decision-making processes. Model architecture and training strategy ResNet-101 is a deep convolutional neural network with 101 layers that enables high-level feature extraction, making it suitable for complex tasks which require capturing detailed spatial hierarchies (He et al. 2016 ). For pose tracking, we chose ResNet-101 to capture the complexity of fish behaviours, as fish move in three-dimensional space, including vertical and depth movements, unlike rodents that primarily move in two dimensions. The pretrained ResNet scaffold requires fewer labeled frames; however, we still labeled approximately 1,000 frames -- more than typically needed for DeepLabCut — to accurately track landmarks. The higher labelling effort was necessary because of fish movements in 3D, the use of a wide-angle lens, and different lighting conditions which introduced variabilities in the video frames. For behaviour classification, we selected the random forest model, as behavioural data (movement, posture, etc.) often involves complex and nonlinear relationships. Additionally, random forests require less data compared to deep learning methods and handle mixed data types and missing values; an advantage given that behavioural datasets typically include both continuous variables (e.g., velocity) and categorical variables (e.g., region of interest) (Tang, 2017 ). We encountered class imbalance issues for behaviours such as swimming, hovering, and resting. For example, hovering was present in approximately 17% of frames and absent in 83% (Fig. 6 a), potentially biasing the classifier toward predicting absence. This imbalance can inflate overall accuracy but result in low recall and missed detections of actual behavioural events. To address this, we employed class weighting, which increases the model’s attention to underrepresented behaviours without duplicating or removing data. Compared to oversampling or under sampling, class weighting retains the complete dataset, is computationally efficient, and avoids the overfitting risks associated with oversampling and the potential data loss from under sampling. Additionally, we opted for a bout-based data split rather than a frame-based split. This method assigns entire behavioural events (bouts) exclusively to either training or testing datasets, preventing time-series leakage and providing a more realistic evaluation of model generalization to new video segments or different individuals. However, this approach may slightly lower reported model performance compared to a frame-based split (Botache, D., 2023). Impact of wide-angle lens distortion on pose estimation and behaviour classification A wide-angle camera lens introduces barrel distortion, an optical effect where straight lines appear curved outward, particularly toward the edges of the frame (Kumar, V. R., 2020). In behaviour tracking, this distortion causes animal body parts to appear unnaturally stretched or bent as they move away from the image center. SimBA uses pose estimation data from DeepLabCut to extract kinematic features (Nilsson, S. R. O, 2020). The wide-angle lens gradually and consistently increases with distance from the frame's center. This radial distortion symmetrically affects points around the center, displacing points further outward as their distance from the center increases. Consequently, body parts closer to the image center remain relatively accurate, while parts near the edges become increasingly distorted. This variability means that kinematic data differ depending on the animal's position within the tank (x, y, and z positions). Such positional dependency introduces significant noise or risk of overfitting, especially for behaviours frequently occurring away from the center (the tank divider), near the edges of the tank. Classifier performance, threshold optimization, and feature interpretability Threshold selection influences classifier performance by balancing sensitivity and specificity. Lower thresholds increase true positive detections but also raise false positives, while higher thresholds reduce false positives at the cost of missed detections. In this study, we selected thresholds that maximized the combined F1-scores for both presence and absence classes. Across all four classifiers, the models were consistently reliable at recognizing when a focal event did not occur. For the negative (absence) class, precision and recall exceeded 85% in every case, peaking at 97% precision for “Not Swim” (Table 2). This strong negative performance establishes a solid baseline for ecological inferences where overestimation of behaviour frequency would be problematic. Performance for the positive (presence) class was more variable and behaviour‑specific. Hunting was detected most effectively, with a precision of 86%, a recall of 96%, and an F1‑score of 0.91 (Table 2a), showing that the feature set captured the distinctive kinematic signature of hunting. Swimming was recall‑driven (85%) but less precise (64%), yielding an F1 of 0.73 (Table 2b); many frames flagged as swimming were indeed true events, yet some of the detections overlapped with other locomotor states. Distance from the prey classification showed an almost perfect F1 score of 96%. (“Far”) achieved a balanced precision of 82% and recall of 66% (F1 = 0.73) (Table 2e), which shows that spatial context derived from pose coordinates is informative but still affected by camera curvature at the tank periphery. By contrast, rest and hover remained challenging: their positive‑class F1‑scores fell to 0.34 and 0.40 (Table 2c, 2d), respectively, driven by both low precision and low recall. In the precision–recall curve for the hunting classifier (Fig. 5 a), at a threshold of 0.00, recall is 100% but precision is only 70% (F1 ≈ 0.82), meaning the model finds every hunting event but also generates some false positives. As the threshold rises to 0.20, precision climbs to 85% while recall remains high at 95% (F1 ≈ 0.90). Between 0.35 and 0.75, the F1 score plateaus around 0.95, with precision at 95–98% and recall at 90–96%, indicating an optimal balance. Above 0.80, recall drops below 80% even as precision nears 100%, causing F1 to decline. The broad plateau of high F1 demonstrates that hunting behaviours occupy a distinct region in feature space: a threshold can be chosen anywhere in the 0.35–0.75 range to favor fewer false negatives or fewer false positives with minimal loss in overall accuracy. Feature-importance analysis (Fig. S1d) shows that key features for detecting hunting behaviour included spatial region-of-interest (ROI) information and repeated use of the feature 'Relative_order_flag_percent'. This suggests that hunting fish often change their vertical posture—for example, by raising the head or curling the tail. Another important feature was 'Center_lag_Tail_dx', which captures horizontal displacement between the body center and the tail. For the swim classifier (Fig. 5 b), at a threshold of 0.00, recall is 99% but precision is only 13% (F1 ≈ 0.23), meaning the model detects nearly all swim bouts yet produces many false positives. As the threshold rises to 0.25, precision improves to 50% and recall remains high at 90% (F1 ≈ 0.64). The optimal balance occurs around 0.45–0.50, where precision and recall both sit at ~ 75%, yielding the peak F1 of 0.75. Beyond 0.60, recall drops below 80% even as precision plateaus at 85–90%, causing F1 to decline; at thresholds > 0.90, recall falls toward zero while precision nears 100%. This steep trade off indicates that swim and non‑swim frames share overlapping kinematic features, so threshold choice drastically shapes the balance between capturing true swims and excluding false positives. Feature-importance ranking (Fig. S1c) shows that 'ROI features' remain highly informative. Other important features include 'Center_vertical_slope_std (240 frames)', which reflects how much the vertical alignment of the body fluctuates over time (240 frames = 8 seconds), and 'Relative_order_flag_percent (240 frames)', which quantifies the proportion of time key body parts (such as the head, midline, and pelvic fin) maintain their expected top-to-bottom anatomical order. This consistent ordering is characteristic of streamlined swimming postures. The feature 'summed_movement_mean' also ranked highly, capturing the overall level of locomotion associated with swimming. For the rest classifier (Fig. 5 d), at a threshold of 0.00, recall is 100% but precision is only 40% (F1 ≈ 0.57), reflecting many false positives. Increasing the threshold to 0.10 raises precision to 66% while recall remains at 90% (F1 ≈ 0.79), marking the optimal balance. Beyond 0.20, recall falls sharply below 45% even as precision hovers around 60%, driving F1 down to ~ 0.34. At very high thresholds (> 0.80), recall approaches zero while precision nears 100%, leaving F1 negligible. This steep decline occurs because resting frames share very similar, low‑motion features with other behaviours, leaving little separability in feature space. However, using that same low threshold of 0.10—while optimal on the original dataset, makes the classifier particularly susceptible to setup variations: when applied to new trials and compared against human expert labels, it yielded a much lower presence F1 (~ 0.34) despite maintaining an absence F1 of 94.4%. Domain shifts in lighting, camera angle, background contrast, and fish positioning changed how resting frames appeared, causing the low cutoff to misclassify many non‑rest frames as resting. Resting was relatively rare, present only ~ 20% of frames (Fig. 6 a), so the labeled training data likely underrepresented the full range of resting poses. Re‑tuning the threshold on a representative set of human‑labeled frames from the new experimental conditions is therefore necessary to recover an overall F1 closer to the originally observed 0.79. Most of the poor rest classifier performance can be explained by the behaviour happening most of the time in the far or mid zones (Fig. 6 b), which has a much greater amplitude of the kinematic variations and pose estimation data are not reliable. For the hover classifier (Fig. 5 c), at a threshold of 0.00, recall is 100% but precision is only ~ 8% (F1 ≈ 0.15), reflecting many false positives. Raising the threshold to 0.35 improves precision to ~ 22% while recall falls to ~ 68% (F1 ≈ 0.33) which shows the peak balance. Beyond 0.40, recall drops toward zero even as precision plateaus around 20%, driving F1 back down. This shallow, low‑ceiling curve occurs because hovering frames share very subtle, noisy motion features with non‑hover frames, leaving little separability in feature space. Unlike resting, poor performance of the hovering classifier cannot be mostly explained by a single factor. Several factors may explain this poor detectability: the wide-angle camera lens introduced image curvature, the fish were poorly oriented (facing towards or away from the camera), and the current extracted feature set may lack variables that capture the unique kinematics of hovering. Hovering and hunting behaviours shared the highest rate of misclassification (Fig. 5 c). Fish frequently exhibited hunting behaviour as they approached the barrier, but upon encountering the obstruction and noticing its presence, they often paused and hovered near it, a behaviour visually very similar to hunting which posed a significant challenge for classification. The difficulty was further compounded by the presence of the fixtures used to hold the barrier, which frequently blocked the fish’s head when individuals contacted the structure. This occlusion disrupted consistent head landmark tracking and introduced variability into the pose estimation data. Unlike hunting, which is typically characterized by a defined approach direction and region-of-interest localization, hovering lacked similarly distinctive or consistently extractable features within the current feature extraction file, which reduced the classifier’s ability to distinguish between the two behaviours. Sources of variability Our experimental setup imposed some sources of variability that constrained classifier performance. First, the use of a wide‑angle lens introduced barrel distortion toward the frame edges. Second, recordings were obtained under non-standardized illumination. While fluorescent lights were on at both dawn and midday trials, large windows made midday trials brighter than dawn trials. Also, the position of the tanks in the room allowed some to receive more light while casting shadows on others, so pose‑estimation confidence had to remain robust across markedly different brightness and shadow patterns. Third, changes in camera tilt and height between trials forced the model to generalize across different viewing geometries. Collectively, these lens-induced distortions, lighting inconsistencies, and camera-angle changes increased feature noise and imposed stricter demands on classifier robustness. Ecological implications Accurately quantifying lionfish behaviours is essential for understanding their ecological impacts on reef communities, particularly as these invasive species continue expanding their geographic range and encounter changing environmental conditions driven by climate change. Variations in temperature, prey availability, and competitive interactions can significantly alter lionfish behavioural patterns, affecting predation pressure, habitat selection, and reproductive success. Hence, obtaining precise behavioural data is important for accurately predicting ecological outcomes and informing effective management strategies. Our study addresses the challenges of efficiently analyzing behaviour by introducing a user-friendly machine learning workflow designed specifically to be accessible to ecologists without advanced computational expertise. The pipeline includes a simplified feature extractor with a customizable behaviour list, allowing users to easily add or exclude behaviours without modifying the code. A companion guide helps users select the most relevant kinematic features for each behavioural context. Complicated machine learning workflows necessitate involvement from specialized personnel skilled in computer science, who, despite their technical proficiency, often lack the ecological expertise needed for nuanced behavioural annotation. By contrast, our simplified pipeline empowers ecologists to directly annotate behavioural video data themselves, thereby building accurate and customized classifiers based on their expert understanding of the species. This streamlined process reduces the time and resources traditionally spent training external observers or technical specialists, which enables researchers to expand their studies through increased sample sizes, longer observation periods, and incorporation of additional ecological parameters. Additionally, the modular structure of our workflow allows straightforward adaptation for other medium-bodied fishes such as Asian carps, largemouth bass ( Micropterus nigricans ), and rainbow trout ( Oncorhynchus mykiss ). Ecologists can incorporate their domain-specific knowledge to develop tailored classifiers for precise and relevant behavioural categorization. Moving forward, we plan to further optimize this methodology for future lionfish and invasive species research by incorporating 3-dimensional behavioural analysis using an additional side-view camera, deploying real-time monitoring models directly in field settings, and expanding analytical capabilities to include additional behavioural and ecological metrics. These advancements will enhance ecologists' ability to monitor invasive species and better understand their impacts on ecosystems amid ongoing environmental change. Conclusion Invasive lionfish are remarkably effective predators in the western Atlantic, where they often encounter naive prey. The ability to correctly and efficiently quantify lionfish predatory behaviours, at least in a laboratory setting, can aid researchers in understanding this invader’s incredible success. We paired DeepLabCut pose‑tracking with SimBA random‑forest classifiers to score four behaviours: hovering, resting, swimming and hunting, in 12 h of invasive lionfish footage. The model reliably detects both the presence and absence of swimming and hunting, whereas hovering and resting are harder to flag when they occur, even though absence is still predicted well. Wide‑angle optics give accurate pose estimates at the frame centre but introduce edge distortion that negatively affect both pose extraction and behaviour classifications. Future studies could benefit from distortion-free optics and tailored feature-extraction pipelines, ideally separating high-motion behaviours (swimming, hunting) from low-motion behaviours (hovering, resting), while a fixed front-view camera with a standard lens, stable high-quality lighting, and a high-contrast uniform background should further reduce pose-estimation noise and boost classification accuracy across all behaviour types. Declarations Acknowledgments We would like to thank Drs. P. Peres-Neto and G. Brown for their advice and feedback on an earlier version of this manuscript, and Tafari J. Smith, who helped with acquiring and processing data. This work was supported by NSERC (Grant No. 2024-05091). Applying machine learning tools for automated behaviour classification in invasive lionfish and comparison with human observations Compliance with Ethical Standards This work was supported by NSERC (Grant numbers 2024-05091). The authors have no relevant financial or non-financial interests to disclose. All behavioural trials were conducted under an animal care permit (#30019689) from Concordia University, Quebec, Canada. Data Availability The analysis code, scripts and supplementary figures and tables in this study are publicly archived on Zenodo at https://doi.org/10.5281/zenodo.15734546 References Albins MA, Hixon MA (2008) Invasive Indo-Pacific lionfish Pterois volitans reduce recruitment of Atlantic coral-reef fishes. Marine Ecology Progress Series 367:233–238. https://doi.org/10.3354/meps07620. Bernardi G, Azzurro E, Bariche M et al. (2024) Invasion genomics of lionfish in the Mediterranean Sea. Ecology and Evolution 14:e11087. https://doi.org/10.1002/ece3.11087. Christin S, Hervet É, Lecomte N (2019) Applications for deep learning in ecology. Methods in Ecology and Evolution 10:1632–1644. https://doi.org/10.1111/2041-210X.13256. Côté IM, Smith NS (2018) The lionfish Pterois sp. invasion: Has the worst-case scenario come to pass? Journal of Fish Biology 92:660–689. https://doi.org/10.1111/jfb.13544. Chanthongdee K, Fuentealba Y, Wahlestedt T et al. (2024) Comprehensive ethological analysis of fear expression in rats using DeepLabCut and SimBA machine learning model. Frontiers in Behavioral Neuroscience 18:1440601. https://doi.org/10.3389/fnbeh.2024.1440601. Green SJ, Dilley ER, Benkwitt CE et al. (2019) Trait-mediated foraging drives patterns of selective predation by native and invasive coral-reef fishes. Ecosphere 10:e02752. https://doi.org/10.1002/ecs2.2752. Green SJ, Akins JL, Maljković A, Côté IM (2012) Invasive lionfish drive Atlantic coral-reef fish declines. PLoS One 7:e32596. https://doi.org/10.1371/journal.pone.0032596. He K, Zhang X, Ren S, Sun J (2016) Deep residual learning for image recognition. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pp 770–778. https://doi.org/10.1109/CVPR.2016.90. Huang N, Lu G, Xu D (2016) A permutation importance-based feature selection method for short-term electricity load forecasting using random forest. Energies 9:767. https://doi.org/10.3390/en9100767. Hsu AI, Yttri EA (2021) B-SOiD: An open-source unsupervised algorithm for identification and fast prediction of behaviours. Nature Communications 12:5188. https://doi.org/10.1038/s41467-021-25420-x. IPBES (2023) Summary for policymakers of the thematic assessment report on invasive alien species and their control. Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services, Bonn. Mathis A, Mamidanna P, Cury KM et al. (2018) DeepLabCut: Markerless pose estimation of user-defined body parts with deep learning. Nature Neuroscience 21:1281–1289. https://doi.org/10.1038/s41593-018-0209-y. Morris JA Jr, Akins JL (2009) Feeding ecology of invasive lionfish (Pterois volitans) in the Bahamian archipelago. Environmental Biology of Fishes 86:389–398. https://doi.org/10.1007/s10641-009-9538-8. Nilsson SRO, Goodwin NL, Choong JJ et al. (2020) Simple Behavioural Analysis (SimBA): An open-source toolkit for computer classification of complex social behaviors in experimental animals. bioRxiv [pre-print]. https://doi.org/10.1101/2020.04.19.049452. Pereira TD, Aldarondo DE, Willmore L et al. (2019) Fast animal pose estimation using deep neural networks. Nature Methods 16:117–125. https://doi.org/10.1038/s41592-018-0234-5. Soares MO, Pereira PHC, Feitosa CV et al. (2023) Lessons from the invasion front: Integration of research and management of the lionfish invasion in Brazil. Journal of Environmental Management 340:117954. https://doi.org/10.1016/j.jenvman.2023.117954. Savitzky A, Golay MJE (1964) Smoothing and differentiation of data by simplified least-squares procedures. Analytical Chemistry 36:1627–1639. https://doi.org/10.1021/ac60214a047. Tang F, Ishwaran H (2017) Random forest missing-data algorithms. Statistical Analysis and Data Mining 10:363–377. https://doi.org/10.1002/sam.11348. Weber RZ, Mulders G, Kaiser J et al. (2022) Deep learning-based behavioral profiling of rodent stroke recovery. BMC Biology 20:232. https://doi.org/10.1186/s12915-022-01434-9. Zhou Y, Liu Y, Li Z et al. (2021) Random forest model based on particle swarm optimization for prediction of surface roughness. Procedia Computer Science 184:393–401. https://doi.org/10.1016/j.procs.2021.03.050. Cite Share Download PDF Status: Published Journal Publication published 16 Feb, 2026 Read the published version in Marine Biology → Version 1 posted Editorial decision: Acceptable after minor revision 16 Sep, 2025 Reviewers agreed at journal 25 Jul, 2025 Reviewers invited by journal 25 Jul, 2025 Editor assigned by journal 23 Jul, 2025 First submitted to journal 22 Jul, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7153188","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":490628831,"identity":"fd9372ab-a651-48ce-8ffc-1679eb08ab55","order_by":0,"name":"Amirreza Khodaparast Kelidbari","email":"data:image/png;base64,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","orcid":"https://orcid.org/0009-0002-2991-0860","institution":"McGill University","correspondingAuthor":true,"prefix":"","firstName":"Amirreza","middleName":"Khodaparast","lastName":"Kelidbari","suffix":""},{"id":490628832,"identity":"95d73051-f5c2-4836-b8c6-4b0cefc74213","order_by":1,"name":"Katelyn Moffat","email":"","orcid":"","institution":"Concordia University","correspondingAuthor":false,"prefix":"","firstName":"Katelyn","middleName":"","lastName":"Moffat","suffix":""},{"id":490628833,"identity":"fae41eb8-2f8c-442b-ab23-b50ebe3245e8","order_by":2,"name":"Iris George","email":"","orcid":"","institution":"Concordia University","correspondingAuthor":false,"prefix":"","firstName":"Iris","middleName":"","lastName":"George","suffix":""},{"id":490628834,"identity":"827df3c4-b188-4696-9661-23babd296ac9","order_by":3,"name":"Nicola Smith","email":"","orcid":"","institution":"Concordia University","correspondingAuthor":false,"prefix":"","firstName":"Nicola","middleName":"","lastName":"Smith","suffix":""}],"badges":[],"createdAt":"2025-07-18 02:50:24","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7153188/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7153188/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s00227-025-04777-3","type":"published","date":"2026-02-16T15:57:02+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":87723822,"identity":"7e96309f-c9f1-4486-ba44-5d819bc458a2","added_by":"auto","created_at":"2025-07-28 10:23:14","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":121947,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAutomated behavioural-analysis pipeline.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eVideo recordings are first acquired and processed. DeepLabCut provides markerless pose estimation, and the resulting pose data are fed into SimBA’s supervised learning module for behaviour detection and classification. The machine learning-generated labels are then compared to human observations for validation\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7153188/v1/0fe92cdd00e2823040bc6079.png"},{"id":87723812,"identity":"e720161a-62e3-493c-9e2f-7e1150a92244","added_by":"auto","created_at":"2025-07-28 10:23:13","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":572650,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eExperimental tank setup\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe predator (invasive lionfish) was separated from the prey (native parrotfish) by both transparent dividers (screen mesh) and an opaque divider (black sponge). At the start of the experiment, the opaque divider was removed so that the fishes could see but not physically contact each other, and lionfish behaviors were recorded. An ultra wide camera captured the entire tank setup. The black grid in the background served as a spatial reference for estimating lionfish position and movement.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7153188/v1/0902871a26fd9f64e12721c1.png"},{"id":87724837,"identity":"263f29da-a95a-4f23-bb8b-4169c39a9308","added_by":"auto","created_at":"2025-07-28 10:31:14","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":153963,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDetection likelihood of annotated body parts in DeepLabCut.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eLionfish body parts with distinctive shapes and colours, such as the head and certain fin tips, showed high detectability in DeepLabCut. In contrast, some spines and fins had lower detectability due to frequent overlapping and the lack of uniquely identifiable features. The dorsal spines, however, showed relatively higher detection likelihood and may be more accurately tracked using an optimized setup. The mid-body region was often covered by the pectoral fins, resulting in consistently low detection likelihood.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7153188/v1/1f95fa2c2fb7f8901ae827eb.png"},{"id":87723797,"identity":"baf0c7dc-3f33-43cd-a61d-cb4a8d3149f5","added_by":"auto","created_at":"2025-07-28 10:23:12","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":266980,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eZone division in the experimental tank.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe experimental tank was divided into five zones based on vertical and horizontal banding, providing structured regions to analyze lionfish movement and positioning. The \"bottom\" zone spans the entire lower row across the tank, with \"sub-bottom\" defined as half of that band. The \"near\" zone includes the first two vertical bands adjacent to the transparent divider, while the \"mid\" zone extends from vertical band two through band five. The top zone covers any area above the first horizontal band. The \"far\" zone covers vertical bands six through eight.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-7153188/v1/b69bfbe253caff571a22b070.png"},{"id":87723803,"identity":"4cb966aa-f71f-4f11-bf43-013935aa8e8e","added_by":"auto","created_at":"2025-07-28 10:23:13","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":359821,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eExternal validation performance heatmaps.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eClassifier performance heatmaps for the five classifications: hunting, distance from prey, swimming, resting and hovering. Machine learning results and human observations are compared. In each panel the top row (“Not”) shows metrics for the absence class and the bottom row for the presence class. Precision is the proportion of predicted positives that are correct, recall is the proportion of true positives detected, and F1-score is their harmonic mean.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-7153188/v1/0e5bc1220be9eb4cd94e43e8.png"},{"id":87723792,"identity":"d77f7c0b-e47d-43a1-864d-748b1b20ce1d","added_by":"auto","created_at":"2025-07-28 10:23:12","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":195185,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eInternal cross-validation performance heatmaps\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePerformance heatmaps from SimBA’s automated classifier evaluation for four behaviors: hunting, swimming, resting, and hovering. In each panel, the upper row (“Not”) reports absence metrics and the lower row reports presence metrics. Precision is the proportion of predicted positives that are correct, recall is the proportion of true positives detected, and F1-score is their harmonic mean\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-7153188/v1/c18f435d4fe37ffa39ae725b.png"},{"id":87723838,"identity":"75daac84-483c-4ed8-9417-40db39501f8d","added_by":"auto","created_at":"2025-07-28 10:23:15","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":195320,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eClassifier performance analysis by precision-recall curve across behaviours.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePrecision–recall curves plot model performance across decision thresholds. The curves reveal how changes in threshold affect the trade-off between false positives (precision) and false negatives (recall).\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-7153188/v1/612c0fac81ba3db1e1f7e29d.png"},{"id":87723815,"identity":"65f64ebd-fc8c-4e29-aefc-298ad7027cf8","added_by":"auto","created_at":"2025-07-28 10:23:14","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":83525,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eBehaviour distribution, spatial bias, and misclassification patterns.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA: overall distribution of annotated behaviours according to human observation, which shows class imbalance of absence and presence of lionfish behaviours; B: illustrates the behaviour composition in \"far\" regions (oi.e., frame edges), where lens distortion is strongest; C: shows which behaviours were mistakenly predicted when the ground truth was \"hover\".\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-7153188/v1/ce3bc5bb86ccdf222b94a0ea.png"},{"id":103251026,"identity":"1a6e025f-aeaa-4424-9492-3077f8853c6e","added_by":"auto","created_at":"2026-02-23 16:01:30","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2995778,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7153188/v1/e6d78e31-b6ae-42b4-97ac-00dee606bfe7.pdf"}],"financialInterests":"","formattedTitle":"Applying machine learning tools for automated behaviour classification in invasive lionfish and comparison with human observations","fulltext":[{"header":"Introduction","content":"\u003cp\u003e\u003cstrong\u003eInvasive lionfish and its ecological impacts\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA particularly successful pair of marine invaders are the Indo-Pacific lionfishes (\u003cem\u003ePterois volitans\u003c/em\u003e and \u003cem\u003eP. miles\u003c/em\u003e), which have colonized the western Atlantic Ocean. Introduced via the aquarium trade in the mid-1980s, lionfish were first documented along Florida\u0026rsquo;s east coast in the early 1990s. By the early 2000s, they had rapidly established populations along the U.S. eastern seaboard and subsequently spread throughout the Caribbean and Gulf of Mexico. Lionfish are now established as far north as North Carolina in the US and as far south as Brazil (Morris \u0026amp; Akins, \u003cspan class=\"CitationRef\"\u003e2009\u003c/span\u003e; Soares et al. \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). They are also currently invading the eastern Mediterranean (Bernardi et al. \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e). Able to thrive in diverse habitats, from estuaries and shallow coral reefs to mesophotic reefs at depths exceeding 200 meters, lionfish have demonstrated remarkable adaptability, making their invasion one of the most rapid and geographically extensive marine fish invasions documented to date (Green et al. \u003cspan class=\"CitationRef\"\u003e2012\u003c/span\u003e; Albins \u0026amp; Hixon, \u003cspan class=\"CitationRef\"\u003e2008\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003eLionfish feed on up to 70 species of native reef fishes and numerous crustaceans in their invaded range, many of high ecological and/or commercial value (C\u0026ocirc;t\u0026eacute; and Smith, \u003cspan class=\"CitationRef\"\u003e2018\u003c/span\u003e). They use an ambush strategy characterized by slow stalking, sudden strikes, and flared pectoral fins to corner prey (Green et al. \u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e). This technique has been highly effective in the Atlantic, where na\u0026iuml;ve prey have not evolved defenses against this novel predator. In some regions, such as The Bahamas, lionfish predation has led to 65% declines in native prey fish biomass on natural reefs within two years of establishment (Green et al. \u003cspan class=\"CitationRef\"\u003e2012\u003c/span\u003e). Fish reductions have, in some cases, resulted in localized extirpations of vulnerable prey species and the replacement of native mesopredator biomass on some reefs (C\u0026ocirc;t\u0026eacute; \u0026amp; Smith, \u003cspan class=\"CitationRef\"\u003e2018\u003c/span\u003e). Notably, field surveys in The Bahamas have observed lionfish occupying the ecological niche once held by native groupers, with uncertain consequences for reef communities (C\u0026ocirc;t\u0026eacute; \u0026amp; Smith, \u003cspan class=\"CitationRef\"\u003e2018\u003c/span\u003e). Given the substantial, negative ecological impacts of lionfish, accurately quantifying their predatory behaviours is essential for effective invasive species management and coral reef conservation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMachine learning in behavioural analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eModern neuroscience and ecology are increasingly leveraging machine learning (ML) techniques for automated animal behaviour tracking and classification (Christin et al. \u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e). Traditionally, researchers depended on manual observation and scoring of behaviours, an approach that is labour-intensive, requires extensive training, and can be prone to observer bias (Mathis et al. \u003cspan class=\"CitationRef\"\u003e2018\u003c/span\u003e). By contrast, ML-based approaches offer a high-throughput alternative, enabling large volumes of video data to be processed while maintaining consistent scoring standards at reduced costs (Nilsson et al. \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003eTwo popular open-source ML tools for automated behavioural analysis are DeepLabCut and SimBA. DeepLabCut uses deep neural networks to estimate animal poses in a markerless fashion, enabling precise tracking of body parts (Mathis et al. \u003cspan class=\"CitationRef\"\u003e2018\u003c/span\u003e). SimBA (Simple Behavioural Analysis) expands on such pose data with a user-friendly platform for training supervised classifiers to detect specific behaviours using random forest models (Nilsson et al. \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e). Random forests, which are ensembles of decision trees trained on random subsets of features and data, effectively mitigate overfitting that single decision trees often face (Zhou, 2021). Both tools have robust user communities that offer support for setup, troubleshooting, and customization. Although DeepLabCut and SimBA have been used and validated in many studies involving rodents (Weber et al. \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e; Chanthongdee., 2024), relatively few have tested these tools on medium-bodied fishes, including invasive lionfish (Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStudy objectives\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn this study, we conduct a controlled experiment using an aquarium setup to evaluate the effectiveness of a machine learning workflow for detecting and quantifying behaviours in mid-bodied fishes, specifically invasive lionfish. Our approach combines deep neural networks for pose estimation (via DeepLabCut) with random forest classifiers (using SimBA) to automatically classify four behaviours: hovering, resting, swimming, and hunting. We aim to address two questions. First, can machine learning methods accurately quantify lionfish behaviours using a wide-angle lens? Second, how does the performance of our automated classification compare to human observations?\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cstrong\u003eExperimental setup and lionfish behavioural trials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBehavioural trials occurred twice a day, at dawn and midday, for two days during July\u0026ndash;September 2024 at the Cape Eleuthera Institute on Eleuthera Island, The Bahamas. Each trial consisted of a 75 G aquarium tank that was separated into two sections by opaque and transparent dividers (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). An adult lionfish (mean\u0026thinsp;\u0026plusmn;\u0026thinsp;se\u0026thinsp;=\u0026thinsp;20.38 cm\u0026thinsp;\u0026plusmn;\u0026thinsp;1.43 total length, TL) was placed in one part of the tank while three to four juvenile, native prey fishes (mean\u0026thinsp;\u0026plusmn;\u0026thinsp;se\u0026thinsp;=\u0026thinsp;6.24 cm\u0026thinsp;\u0026plusmn;\u0026thinsp;0.21 TL) (princess and striped parrotfish, \u003cem\u003eScarus taeniopterus\u003c/em\u003e and \u003cem\u003eS. iseri\u003c/em\u003e) were placed in the other part. Fish shelters in the form of rocks and PVC pipe were also placed in each section of the tank (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). At the start of a trial, the opaque divider was removed while the transparent divider remained in place. This setup allowed the fishes to see one another but prevented physical contact. Once a trial began, 30 minutes of video footage of lionfish behaviours was recorded using GoPro Hero11 cameras in wide-angle mode, set to 4K resolution at 30 frames per second. Footage was then downscaled to 1080p with FFmpeg for computational efficiency. All video files were labelled and stored, totaling approximately 12 hours of footage of 7 adult lionfish. Lionfish were never reused in the experiment except between dawn and mid-day trials. All behavioural trials were conducted under an animal care permit (#30019689) from Concordia University, Quebec, Canada.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eHuman observations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTwo human observers were trained to classify a pre-defined set of lionfish behaviours and record lionfish position relative to the dividers (i.e., near, mid, or far) at 30-second intervals for each 30-minute trial, across 24 total trials, resulting in 1,219 data points for comparison with machine generated predictions. Observers officially recorded their data once there was at least 95% agreement between them (inter-observer reliability) and a repeatability score of 95% within each observer (intra-observer reliability). To confirm repeatability, each observer rewatched and rescored a portion of lionfish footage until consistency in behaviour scoring was achieved across multiple assessments.\u003c/p\u003e\n\u003cp\u003eThe predator (invasive lionfish) was separated from the prey (native parrotfish) by both transparent dividers (screen mesh) and an opaque divider (black sponge). At the start of the experiment, the opaque divider was removed so that the fishes could see but not physically contact each other, and lionfish behaviors were recorded. An ultra wide camera captured the entire tank setup. The black grid in the background served as a spatial reference for estimating lionfish position and movement.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDeepLabCut pose estimation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDeepLabCut (Mathis et al. \u003cspan class=\"CitationRef\"\u003e2018\u003c/span\u003e) was used to track lionfish movements (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e). A total of 1,000 video frames were extracted from pilot recordings captured in wide-angle mode, which introduced greater variation in lionfish size, shape, and orientation across the frame. While\u0026thinsp;~\u0026thinsp;300 frames are typically sufficient for training, the increased visual variability caused by the wide-angle perspective required a larger set of annotated frames to ensure the model could accurately generalize across diverse poses and viewing angles.\u003c/p\u003e\n\u003cp\u003eFrame annotation was performed in Napari to label specific anatomical points on each frame (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e). A ResNet-101 backbone was then trained on a cloud-based NVIDIA L4 GPU for four hours, using DeepLabCut\u0026rsquo;s default training parameters (including max_input_size: 1500, min_input_size: 64, global_scale: 0.8, pos_dist_thresh: 17, pafwidth: 20, scale_jitter_lo: 0.5, scale_jitter_up: 1.25) and adjusting only the batch size to accommodate GPU memory constraints. The performance of the pose estimation model was evaluated following training with DeepLabCut for 30,000 iterations, at which point the test error had stabilized. Of the annotated frames, 95% were allocated to the training set, while the remaining 5% were used for testing (shuffle 1). With images recorded at 1 920 \u0026times; 1 080 px, the network achieved an average training error of 6.21 px and a test error of 5.32 px (p-cutoff\u0026thinsp;=\u0026thinsp;0.65).\u003c/p\u003e\n\u003cp\u003eThe \u0026lt;\u0026thinsp;1-px train\u0026ndash;test gap meets the DeepLabCut guideline that comparable errors indicate good generalisation (Mathis et al. \u003cspan class=\"CitationRef\"\u003e2018\u003c/span\u003e). In absolute terms, 5.32 px corresponds to 0.28% of frame width \u0026mdash; within the 2\u0026ndash;6-px range reported for benchmark rodent and fish datasets (Mathis et al. \u003cspan class=\"CitationRef\"\u003e2018\u003c/span\u003e; Pereira et al. \u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e). These results showed that the model was able to accurately predict body part locations when the fish has the optimal orientation (side view) towards the camera. The trained network was then used to analyze additional lionfish videos, again employing the NVIDIA L4 GPU to ensure efficient inference.\u003c/p\u003e\n\u003cp\u003eLionfish body parts with distinctive shapes and colours, such as the head and certain fin tips, showed high detectability in DeepLabCut. In contrast, some spines and fins had lower detectability due to frequent overlapping and the lack of uniquely identifiable features. The dorsal spines, however, showed relatively higher detection likelihood and may be more accurately tracked using an optimized setup. The mid-body region was often covered by the pectoral fins, resulting in consistently low detection likelihood.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSimBA workflow\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCSV pose data extracted from DeepLabCut were imported into the SimBA platform for subsequent behavioural analysis. All pose-estimation outputs were first preprocessed to remove outliers by applying a movement criterion of 1.25 body‐lengths per frame (distance between body parts 1 [mouth] and 10 [caudal fin tip]). This threshold was chosen based on pilot analyses in which we systematically varied the movement criterion from 0.7 BL (body length)/ frame (SimBA\u0026rsquo;s default movement criterion) to 1.5 BL/frame. Thresholds\u0026thinsp;\u0026lt;\u0026thinsp;1.0 BL/frame excluded genuine fast‐start and strike sequences, whereas thresholds\u0026thinsp;\u0026gt;\u0026thinsp;1.5 BL/frame passed most false jumps. At 1.25 BL/frame we achieved the optimal balance to remove the outliers while maintaining genuine rapid movements. Positional trajectories were then smoothed using a Savitzky\u0026ndash;Golay kernel (Savitzky, A. et al. 1964).\u003c/p\u003e\n\u003cp\u003eBased on each frame\u0026rsquo;s body-part coordinates, along with the known frame rate and pixels-per-millimetre ratio, SimBA\u0026rsquo;s feature extraction step generated a set of kinematic and morphological metrics. A modified version of the default zebrafish extraction file available on the SimBA website was used (Table S1; \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5281/zenodo.15734546\u003c/span\u003e\u003c/span\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e)\u003c/span\u003e to accommodate additional lionfish-specific key points. Selecting features via permutation importance before retraining a Random Forest significantly improved short-term load-forecasting performance compared to using all variables (Huang, N. et al. \u003cspan class=\"CitationRef\"\u003e2016\u003c/span\u003e). To evaluate the utility of additional features, we manually added them to the feature extraction file and assessed their contribution using the feature importance rankings in SimBA. Features that failed to improve behavioral differentiation were removed, while those that improved class separation were retained, having ranked within the top 200 most informative features out of 893 in at least one behaviour extracted, based on permutation importance scores. The platform computed features such as rotation/orientation features, path metrics, pairwise distances, measures of velocity and acceleration, and more, resulting in 893 extracted features. SimBA\u0026rsquo;s integrated ROI definition tool was then employed to partition the experimental tank into spatial zones (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e), and these zone assignments were appended to the extracted features to link behaviour with location.\u003c/p\u003e\n\u003cp\u003eA two-hours subset of pilot video footage was manually annotated in SimBA (Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e) to label the presence or absence of four defined behaviours. Kinematic features extracted from pose data were used as input for a supervised machine learning classifier. A random forest model was trained using the following parameters: 2,000 estimators, max_features\u0026thinsp;=\u0026thinsp;sqrt, criterion\u0026thinsp;=\u0026thinsp;entropy, test_size\u0026thinsp;=\u0026thinsp;0.3, and train-test split type\u0026thinsp;=\u0026thinsp;FRAMES. The min_samples_leaf was set to 1, with no minimum or maximum under sampling. Balanced class-weight adjustments were used to account for class imbalances in the presence and absence of resting, hunting, hovering, and swimming behaviours. Classification thresholds were manually determined using the precision-recall curve and the interactive probability vs. frame number plot to find the best balance between precision and recall. The trained model was subsequently applied to additional lionfish csv files to generate behaviour classifications. Fish position was estimated using the weighted average of selected body parts, with weights based on their detection probabilities.\u003c/p\u003e\n\u003cp\u003eThe experimental tank was divided into five zones based on vertical and horizontal banding, providing structured regions to analyze lionfish movement and positioning. The \u0026quot;bottom\u0026quot; zone spans the entire lower row across the tank, with \u0026quot;sub-bottom\u0026quot; defined as half of that band. The \u0026quot;near\u0026quot; zone includes the first two vertical bands adjacent to the transparent divider, while the \u0026quot;mid\u0026quot; zone extends from vertical band two through band five. The top zone covers any area above the first horizontal band. The \u0026quot;far\u0026quot; zone covers vertical bands six through eight.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003cp\u003e\u003cstrong\u003eTable 1. Behavioural classifiers and operational definitions for lionfish.\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eDefinitions of four distinct lionfish behaviours: hovering, resting, swimming, and hunting.\u003c/p\u003e\n \u003cp\u003e\u003cimg src=\"data:image/png;base64,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\"\u003e\u003c/p\u003e\n \u003cdiv align=\"left\" class=\"colspec\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003cdiv align=\"left\" class=\"colspec\"\u003eTo assess the performance of the behavioural classifiers, we computed three standard evaluation metrics in the absence and presence of the behaviours: recall, precision, and F1 score.\u003c/div\u003e\n\u003c/div\u003e\n\u003cp\u003ea. Recall (absence) - measures how well the classifier correctly identifies when the behaviour is truly absent.\u003c/p\u003e\n\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e$$\\:Recall\\left(absence\\right)\\:=\\frac{True\\:Negatives}{True\\:Negatives\\:+\\:False\\:Positives}$$\u003c/div\u003e\n\u003c/div\u003e\n\u003cp\u003eb. Recall (presence) - measures how well the classifier detected the behaviour when it was truly present.\u003c/p\u003e\n\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e$$\\:Recall\\left(presence\\right)=\\frac{True\\:Positives}{True\\:Positives\\:+\\:False\\:Negatives}\\:$$\u003c/div\u003e\n\u003c/div\u003e\n\u003cp\u003e2a. Precision (absent) - indicates how often a prediction of absence was correct.\u003c/p\u003e\n\u003cdiv id=\"Equc\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equc\" name=\"EquationSource\"\u003e$$\\:\\:Precision\\left(absent\\right)=\\:\\frac{True\\:Negatives}{True\\:Negative\\:+False\\:Negative}$$\u003c/div\u003e\n\u003c/div\u003e\n\u003cp\u003e2b. Precision (present) - indicates how often the classifier was correct when it predicted that a behaviour was present.\u003c/p\u003e\n\u003cdiv id=\"Equd\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equd\" name=\"EquationSource\"\u003e$$\\:Precision\\left(present\\right)\\:=\\frac{True\\:Positive}{True\\:Positive\\:+\\:False\\:Positive}\\:$$\u003c/div\u003e\n\u003c/div\u003e\n\u003cp\u003e3. F1 score - provides a single metric that balances precision and recall.\u003c/p\u003e\n\u003cdiv id=\"Eque\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Eque\" name=\"EquationSource\"\u003e$$\\:F1=\\:\\frac{2\\:\\times\\:\\:Precision\\:\\times\\:\\:Recall}{Precision\\:+\\:Recall}$$\u003c/div\u003e\n\u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eHuman agreement and reference set validation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTwo human observers independently annotated the four behaviours: hovering, resting, hunting, and swimming, with both inter-observer agreement and intra-observer repeatability\u0026thinsp;\u0026ge;\u0026thinsp;95%. For spatial classification (near or far zones), agreement reached 99% for both inter- and intra-observer comparisons.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eExternal validation: classifiers vs. human observations consensus\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eExternal validation measures how well the trained classifier reproduces labels made by independent human observers on truly unseen data and setups. For each behaviour we built a confusion matrix by tallying true negatives (TN), false negatives (FN), true positives (TP) and false positives (FP; Table S2). Using the confusion matrices (Table S1), precision, recall and F1 were separately calculated for the presence and absence of that behaviour. Classifier performance differed across behaviours (Table 2). Specifically for hunting, the model correctly identified 694 of 722 true events (Table S2), yielding 86.2% precision and 96.1% recall (F1\u0026thinsp;=\u0026thinsp;0.91) (Table 2a). The negative class was also reliable, with 92.9% precision and 77.7% recall, indicating relatively few false alarms (Table 2a). Swimming showed a similar profile: recall remained high (85.3%), capturing most swim bouts, but precision was moderate (64.1%) and showed some mislabelling; the resulting F1score was 73% (Table 2b). In contrast, rest and hover were not as reliable. Rest was detected in only 34.8% of true instances and with 33.0% precision (F1\u0026thinsp;=\u0026thinsp;0.34) (Table 2d), while hover achieved 30.6% recall and 57.8% precision (F1\u0026thinsp;=\u0026thinsp;0.40) (Table 2c). In both cases, precision and recall for the absence class exceeded 84% (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003ec \u0026minus;\u0026thinsp;2d). For distance from the prey estimation, when the fish was near the prey, the location was rarely misclassified with 95.0% precision, 96.0% recall (Table 2e). When the fish was in the mid or far zones (Fig. \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e), its location was detected only two‑thirds of the time, with 82.0% precision, 66.0% recall, and F1\u0026thinsp;=\u0026thinsp;0.73 (Table 2e).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInternal cross‑validation: model generalisation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWithin SimBA, we withheld 30% of our author‑labelled frames, ensuring whole behavioural bouts remained intact to prevent temporal leakage, and trained on the remaining 70%. SimBA then automatically generated per‑behaviour confusion matrices (TN, FP, FN, TP) on this internal test split and calculated precision, recall and F1 for both presence and absence of each behaviour.\u003c/p\u003e\n\u003cp\u003eHunting (Table\u0026nbsp;3a) was detected with high fidelity: presence precision reached 97.7% and recall 90.8% (F1\u0026thinsp;=\u0026thinsp;94.1%), while the absence class also performed strongly (82.5% precision, 95.4% recall; F1\u0026thinsp;=\u0026thinsp;88.5%), indicating very few false positives. Swimming (Table\u0026nbsp;3b) exhibited a similarly robust absence profile\u0026mdash;98.1% precision and 96.9% recall (F₁ = 97.5%) and a good presence performance (73.3% precision, 81.8% recall; F1\u0026thinsp;=\u0026thinsp;77.3%), with some misclassifications of actual swim bouts. In contrast, resting and hovering were far less reliable. Rest events were flagged with just 49.4% precision and 25.2% recall (F1\u0026thinsp;=\u0026thinsp;33.4%) (Table\u0026nbsp;3d), while hover classifier achieved only 21.8% precision and 56.3% recall (F1\u0026thinsp;=\u0026thinsp;31.4%) (Table\u0026nbsp;2c).\u003c/p\u003e\n\u003cp\u003eDespite these low presence scores, absence metrics for both behaviours remained high (hover: 98.1% precision, 91.9% recall; rest: 82.5% precision, 93.1% recall). The close agreement between F1 values of internal and external validations indicates that our pilot training videos effectively captured nearly the full diversity of behavioural poses encountered in independent datasets.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eThreshold analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe also looked at classifier performance by the precision-recall curve across behaviours. A precision\u0026ndash;recall curve plots how precision and recall trade off as the threshold is changed in a classifier. The classifier for hunting behaviour showed excellent performance across thresholds, with both precision and recall remaining consistently high (Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003ea). The F1 score peaked around ~\u0026thinsp;0.5, with a strong balance between detecting true hunting events and minimizing false positives (Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003ea). In contrast, the swim classifier showed a clear precision-recall trade off, with precision increasing and recall decreasing as the threshold rose (Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003eb). The F1 score peaked around a threshold of 0.55 at a\u0026thinsp;~\u0026thinsp;0.75 performance, which indicates good overall performance (Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003eb). The resting behaviour classifier exhibited a steep precision\u0026ndash;recall trade off driven by overlapping kinematic features with non‑rest frames (Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003ed). At a discrimination threshold of 0.00, the model achieved maximal recall (100%) but low precision (\u0026asymp;\u0026thinsp;40%), yielding an F1 of 0.57 (Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003ed). Increasing the threshold to 0.10 boosted precision to 66% while maintaining recall at 90%, producing the peak F1 of 0.79 (Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003ed). Beyond a threshold of 0.20, recall fell below 30% and precision plateaued around 50%, causing F1 to decline to ~\u0026thinsp;0.34 (Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003ed). Lastly, the hover classifier showed a very shallow trade off: at a threshold of 0.00 it achieved 100% recall but only\u0026thinsp;~\u0026thinsp;8% precision (F1\u0026thinsp;\u0026asymp;\u0026thinsp;0.15), and at the optimal threshold (~\u0026thinsp;0.35) precision rose to ~\u0026thinsp;22% while recall fell to ~\u0026thinsp;68%, yielding a peak F1 of just\u0026thinsp;~\u0026thinsp;0.33 (Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003ec). Beyond the threshold of 0.40, recall continued to decline toward zero even as precision plateaued around 20%, driving F1 back down (Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003ec).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eBehaviour composition and misclassification patterns\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBehaviour distribution analysis revealed significant class imbalance, with hunting being the most frequent behaviour observed (Fig. \u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003ea). Spatial analysis revealed that resting occurred mostly at tank edges, areas prone to increased lens distortion (Fig. \u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003eb) while hovering is commonly misclassified as hunting (Fig. \u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003ec).\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eWe combined DeepLabCut\u0026rsquo;s deep learning\u0026ndash;based pose estimation with SimBA\u0026rsquo;s random forest classifiers to detect four lionfish behaviours. The models accurately classified high-motion behaviours (hunting and swimming) achieving high precision and recall. In contrast, low-motion behaviours (hovering and resting) were harder to detect. We also found that while tracking remained reliable near the frame center, distortion at the edges reduced pose accuracy and degraded classification performance.\u003c/p\u003e\u003cp\u003e\u003cb\u003eDiscrete vs continuously variable behaviour\u003c/b\u003e\u003c/p\u003e\u003cp\u003eBecause SimBA assigns each video frame a continuous likelihood score for behavioural categories rather than strictly discrete labels, this workflow is suitable for studying transitional behaviours. Researchers can quantitatively track how an animal\u0026rsquo;s likelihood of exhibiting a certain behaviour gradually increases or fades away, capturing the temporal dynamics inherent in behavioural transitions. For instance, by plotting the likelihood curve through time, one can visualize a fish's hunting likelihood rising incrementally during preparatory motions, declining briefly if the fish hesitates or initiates an alternative action, and then increasing again upon executing a full predatory lunge. This continuous probabilistic output allows investigators to examine transitional states in detail, quantify their similarity to established behaviour categories, and explore ecological hypotheses related to behavioural plasticity and decision-making processes.\u003c/p\u003e\u003cp\u003e\u003cb\u003eModel architecture and training strategy\u003c/b\u003e\u003c/p\u003e\u003cp\u003eResNet-101 is a deep convolutional neural network with 101 layers that enables high-level feature extraction, making it suitable for complex tasks which require capturing detailed spatial hierarchies (He et al. \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). For pose tracking, we chose ResNet-101 to capture the complexity of fish behaviours, as fish move in three-dimensional space, including vertical and depth movements, unlike rodents that primarily move in two dimensions. The pretrained ResNet scaffold requires fewer labeled frames; however, we still labeled approximately 1,000 frames -- more than typically needed for DeepLabCut \u0026mdash; to accurately track landmarks. The higher labelling effort was necessary because of fish movements in 3D, the use of a wide-angle lens, and different lighting conditions which introduced variabilities in the video frames.\u003c/p\u003e\u003cp\u003eFor behaviour classification, we selected the random forest model, as behavioural data (movement, posture, etc.) often involves complex and nonlinear relationships. Additionally, random forests require less data compared to deep learning methods and handle mixed data types and missing values; an advantage given that behavioural datasets typically include both continuous variables (e.g., velocity) and categorical variables (e.g., region of interest) (Tang, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2017\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eWe encountered class imbalance issues for behaviours such as swimming, hovering, and resting. For example, hovering was present in approximately 17% of frames and absent in 83% (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ea), potentially biasing the classifier toward predicting absence. This imbalance can inflate overall accuracy but result in low recall and missed detections of actual behavioural events. To address this, we employed class weighting, which increases the model\u0026rsquo;s attention to underrepresented behaviours without duplicating or removing data. Compared to oversampling or under sampling, class weighting retains the complete dataset, is computationally efficient, and avoids the overfitting risks associated with oversampling and the potential data loss from under sampling.\u003c/p\u003e\u003cp\u003eAdditionally, we opted for a bout-based data split rather than a frame-based split. This method assigns entire behavioural events (bouts) exclusively to either training or testing datasets, preventing time-series leakage and providing a more realistic evaluation of model generalization to new video segments or different individuals. However, this approach may slightly lower reported model performance compared to a frame-based split (Botache, D., 2023).\u003c/p\u003e\u003cp\u003e\u003cb\u003eImpact of wide-angle lens distortion on pose estimation and behaviour classification\u003c/b\u003e\u003c/p\u003e\u003cp\u003eA wide-angle camera lens introduces barrel distortion, an optical effect where straight lines appear curved outward, particularly toward the edges of the frame (Kumar, V. R., 2020). In behaviour tracking, this distortion causes animal body parts to appear unnaturally stretched or bent as they move away from the image center. SimBA uses pose estimation data from DeepLabCut to extract kinematic features (Nilsson, S. R. O, 2020). The wide-angle lens gradually and consistently increases with distance from the frame's center. This radial distortion symmetrically affects points around the center, displacing points further outward as their distance from the center increases. Consequently, body parts closer to the image center remain relatively accurate, while parts near the edges become increasingly distorted. This variability means that kinematic data differ depending on the animal's position within the tank (x, y, and z positions). Such positional dependency introduces significant noise or risk of overfitting, especially for behaviours frequently occurring away from the center (the tank divider), near the edges of the tank.\u003c/p\u003e\u003cp\u003e\u003cb\u003eClassifier performance, threshold optimization, and feature interpretability\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThreshold selection influences classifier performance by balancing sensitivity and specificity. Lower thresholds increase true positive detections but also raise false positives, while higher thresholds reduce false positives at the cost of missed detections. In this study, we selected thresholds that maximized the combined F1-scores for both presence and absence classes. Across all four classifiers, the models were consistently reliable at recognizing when a focal event did not occur. For the negative (absence) class, precision and recall exceeded 85% in every case, peaking at 97% precision for \u0026ldquo;Not Swim\u0026rdquo; (Table\u0026nbsp;2). This strong negative performance establishes a solid baseline for ecological inferences where overestimation of behaviour frequency would be problematic.\u003c/p\u003e\u003cp\u003ePerformance for the positive (presence) class was more variable and behaviour‑specific. Hunting was detected most effectively, with a precision of 86%, a recall of 96%, and an F1‑score of 0.91 (Table\u0026nbsp;2a), showing that the feature set captured the distinctive kinematic signature of hunting. Swimming was recall‑driven (85%) but less precise (64%), yielding an F1 of 0.73 (Table\u0026nbsp;2b); many frames flagged as swimming were indeed true events, yet some of the detections overlapped with other locomotor states. Distance from the prey classification showed an almost perfect F1 score of 96%. (\u0026ldquo;Far\u0026rdquo;) achieved a balanced precision of 82% and recall of 66% (F1\u0026thinsp;=\u0026thinsp;0.73) (Table\u0026nbsp;2e), which shows that spatial context derived from pose coordinates is informative but still affected by camera curvature at the tank periphery. By contrast, rest and hover remained challenging: their positive‑class F1‑scores fell to 0.34 and 0.40 (Table\u0026nbsp;2c, 2d), respectively, driven by both low precision and low recall.\u003c/p\u003e\u003cp\u003eIn the precision\u0026ndash;recall curve for the hunting classifier (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ea), at a threshold of 0.00, recall is 100% but precision is only 70% (F1\u0026thinsp;\u0026asymp;\u0026thinsp;0.82), meaning the model finds every hunting event but also generates some false positives. As the threshold rises to 0.20, precision climbs to 85% while recall remains high at 95% (F1\u0026thinsp;\u0026asymp;\u0026thinsp;0.90). Between 0.35 and 0.75, the F1 score plateaus around 0.95, with precision at 95\u0026ndash;98% and recall at 90\u0026ndash;96%, indicating an optimal balance. Above 0.80, recall drops below 80% even as precision nears 100%, causing F1 to decline. The broad plateau of high F1 demonstrates that hunting behaviours occupy a distinct region in feature space: a threshold can be chosen anywhere in the 0.35\u0026ndash;0.75 range to favor fewer false negatives or fewer false positives with minimal loss in overall accuracy.\u003c/p\u003e\u003cp\u003eFeature-importance analysis (Fig. S1d) shows that key features for detecting hunting behaviour included spatial region-of-interest (ROI) information and repeated use of the feature 'Relative_order_flag_percent'. This suggests that hunting fish often change their vertical posture\u0026mdash;for example, by raising the head or curling the tail. Another important feature was 'Center_lag_Tail_dx', which captures horizontal displacement between the body center and the tail.\u003c/p\u003e\u003cp\u003eFor the swim classifier (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eb), at a threshold of 0.00, recall is 99% but precision is only 13% (F1\u0026thinsp;\u0026asymp;\u0026thinsp;0.23), meaning the model detects nearly all swim bouts yet produces many false positives. As the threshold rises to 0.25, precision improves to 50% and recall remains high at 90% (F1\u0026thinsp;\u0026asymp;\u0026thinsp;0.64). The optimal balance occurs around 0.45\u0026ndash;0.50, where precision and recall both sit at ~\u0026thinsp;75%, yielding the peak F1 of 0.75. Beyond 0.60, recall drops below 80% even as precision plateaus at 85\u0026ndash;90%, causing F1 to decline; at thresholds\u0026thinsp;\u0026gt;\u0026thinsp;0.90, recall falls toward zero while precision nears 100%. This steep trade off indicates that swim and non‑swim frames share overlapping kinematic features, so threshold choice drastically shapes the balance between capturing true swims and excluding false positives.\u003c/p\u003e\u003cp\u003eFeature-importance ranking (Fig. S1c) shows that 'ROI features' remain highly informative. Other important features include 'Center_vertical_slope_std (240 frames)', which reflects how much the vertical alignment of the body fluctuates over time (240 frames\u0026thinsp;=\u0026thinsp;8 seconds), and 'Relative_order_flag_percent (240 frames)', which quantifies the proportion of time key body parts (such as the head, midline, and pelvic fin) maintain their expected top-to-bottom anatomical order. This consistent ordering is characteristic of streamlined swimming postures. The feature 'summed_movement_mean' also ranked highly, capturing the overall level of locomotion associated with swimming.\u003c/p\u003e\u003cp\u003eFor the rest classifier (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ed), at a threshold of 0.00, recall is 100% but precision is only 40% (F1\u0026thinsp;\u0026asymp;\u0026thinsp;0.57), reflecting many false positives. Increasing the threshold to 0.10 raises precision to 66% while recall remains at 90% (F1\u0026thinsp;\u0026asymp;\u0026thinsp;0.79), marking the optimal balance. Beyond 0.20, recall falls sharply below 45% even as precision hovers around 60%, driving F1 down to ~\u0026thinsp;0.34. At very high thresholds (\u0026gt;\u0026thinsp;0.80), recall approaches zero while precision nears 100%, leaving F1 negligible. This steep decline occurs because resting frames share very similar, low‑motion features with other behaviours, leaving little separability in feature space. However, using that same low threshold of 0.10\u0026mdash;while optimal on the original dataset, makes the classifier particularly susceptible to setup variations: when applied to new trials and compared against human expert labels, it yielded a much lower presence F1 (~\u0026thinsp;0.34) despite maintaining an absence F1 of 94.4%. Domain shifts in lighting, camera angle, background contrast, and fish positioning changed how resting frames appeared, causing the low cutoff to misclassify many non‑rest frames as resting. Resting was relatively rare, present only\u0026thinsp;~\u0026thinsp;20% of frames (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ea), so the labeled training data likely underrepresented the full range of resting poses. Re‑tuning the threshold on a representative set of human‑labeled frames from the new experimental conditions is therefore necessary to recover an overall F1 closer to the originally observed 0.79. Most of the poor rest classifier performance can be explained by the behaviour happening most of the time in the far or mid zones (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eb), which has a much greater amplitude of the kinematic variations and pose estimation data are not reliable.\u003c/p\u003e\u003cp\u003eFor the hover classifier (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ec), at a threshold of 0.00, recall is 100% but precision is only\u0026thinsp;~\u0026thinsp;8% (F1\u0026thinsp;\u0026asymp;\u0026thinsp;0.15), reflecting many false positives. Raising the threshold to 0.35 improves precision to ~\u0026thinsp;22% while recall falls to ~\u0026thinsp;68% (F1\u0026thinsp;\u0026asymp;\u0026thinsp;0.33) which shows the peak balance. Beyond 0.40, recall drops toward zero even as precision plateaus around 20%, driving F1 back down. This shallow, low‑ceiling curve occurs because hovering frames share very subtle, noisy motion features with non‑hover frames, leaving little separability in feature space. Unlike resting, poor performance of the hovering classifier cannot be mostly explained by a single factor. Several factors may explain this poor detectability: the wide-angle camera lens introduced image curvature, the fish were poorly oriented (facing towards or away from the camera), and the current extracted feature set may lack variables that capture the unique kinematics of hovering.\u003c/p\u003e\u003cp\u003eHovering and hunting behaviours shared the highest rate of misclassification (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ec). Fish frequently exhibited hunting behaviour as they approached the barrier, but upon encountering the obstruction and noticing its presence, they often paused and hovered near it, a behaviour visually very similar to hunting which posed a significant challenge for classification. The difficulty was further compounded by the presence of the fixtures used to hold the barrier, which frequently blocked the fish\u0026rsquo;s head when individuals contacted the structure. This occlusion disrupted consistent head landmark tracking and introduced variability into the pose estimation data. Unlike hunting, which is typically characterized by a defined approach direction and region-of-interest localization, hovering lacked similarly distinctive or consistently extractable features within the current feature extraction file, which reduced the classifier\u0026rsquo;s ability to distinguish between the two behaviours.\u003c/p\u003e\u003cp\u003e\u003cb\u003eSources of variability\u003c/b\u003e\u003c/p\u003e\u003cp\u003eOur experimental setup imposed some sources of variability that constrained classifier performance. First, the use of a wide‑angle lens introduced barrel distortion toward the frame edges. Second, recordings were obtained under non-standardized illumination. While fluorescent lights were on at both dawn and midday trials, large windows made midday trials brighter than dawn trials. Also, the position of the tanks in the room allowed some to receive more light while casting shadows on others, so pose‑estimation confidence had to remain robust across markedly different brightness and shadow patterns. Third, changes in camera tilt and height between trials forced the model to generalize across different viewing geometries. Collectively, these lens-induced distortions, lighting inconsistencies, and camera-angle changes increased feature noise and imposed stricter demands on classifier robustness.\u003c/p\u003e\u003cp\u003e\u003cb\u003eEcological implications\u003c/b\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eAccurately quantifying lionfish behaviours is essential for understanding their ecological impacts on reef communities, particularly as these invasive species continue expanding their geographic range and encounter changing environmental conditions driven by climate change. Variations in temperature, prey availability, and competitive interactions can significantly alter lionfish behavioural patterns, affecting predation pressure, habitat selection, and reproductive success. Hence, obtaining precise behavioural data is important for accurately predicting ecological outcomes and informing effective management strategies. Our study addresses the challenges of efficiently analyzing behaviour by introducing a user-friendly machine learning workflow designed specifically to be accessible to ecologists without advanced computational expertise. The pipeline includes a simplified feature extractor with a customizable behaviour list, allowing users to easily add or exclude behaviours without modifying the code. A companion guide helps users select the most relevant kinematic features for each behavioural context. Complicated machine learning workflows necessitate involvement from specialized personnel skilled in computer science, who, despite their technical proficiency, often lack the ecological expertise needed for nuanced behavioural annotation. By contrast, our simplified pipeline empowers ecologists to directly annotate behavioural video data themselves, thereby building accurate and customized classifiers based on their expert understanding of the species. This streamlined process reduces the time and resources traditionally spent training external observers or technical specialists, which enables researchers to expand their studies through increased sample sizes, longer observation periods, and incorporation of additional ecological parameters.\u003c/p\u003e\u003cp\u003eAdditionally, the modular structure of our workflow allows straightforward adaptation for other medium-bodied fishes such as Asian carps, largemouth bass (\u003cem\u003eMicropterus nigricans\u003c/em\u003e), and rainbow trout (\u003cem\u003eOncorhynchus mykiss\u003c/em\u003e). Ecologists can incorporate their domain-specific knowledge to develop tailored classifiers for precise and relevant behavioural categorization.\u003c/p\u003e\u003cp\u003eMoving forward, we plan to further optimize this methodology for future lionfish and invasive species research by incorporating 3-dimensional behavioural analysis using an additional side-view camera, deploying real-time monitoring models directly in field settings, and expanding analytical capabilities to include additional behavioural and ecological metrics. These advancements will enhance ecologists' ability to monitor invasive species and better understand their impacts on ecosystems amid ongoing environmental change.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eInvasive lionfish are remarkably effective predators in the western Atlantic, where they often encounter naive prey. The ability to correctly and efficiently quantify lionfish predatory behaviours, at least in a laboratory setting, can aid researchers in understanding this invader\u0026rsquo;s incredible success. We paired DeepLabCut pose‑tracking with SimBA random‑forest classifiers to score four behaviours: hovering, resting, swimming and hunting, in 12 h of invasive lionfish footage. The model reliably detects both the presence and absence of swimming and hunting, whereas hovering and resting are harder to flag when they occur, even though absence is still predicted well. Wide‑angle optics give accurate pose estimates at the frame centre but introduce edge distortion that negatively affect both pose extraction and behaviour classifications. Future studies could benefit from distortion-free optics and tailored feature-extraction pipelines, ideally separating high-motion behaviours (swimming, hunting) from low-motion behaviours (hovering, resting), while a fixed front-view camera with a standard lens, stable high-quality lighting, and a high-contrast uniform background should further reduce pose-estimation noise and boost classification accuracy across all behaviour types.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe would like to thank Drs. P. Peres-Neto and G. Brown for their advice and feedback on an earlier version of this manuscript, and Tafari J. Smith, who helped with acquiring and processing data. This work was supported by NSERC (Grant No. 2024-05091).\u003c/p\u003e\n\u003cp\u003eApplying machine learning tools for automated behaviour classification in invasive lionfish and comparison with human observations\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompliance with Ethical Standards\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by NSERC (Grant numbers 2024-05091).\u003c/p\u003e\n\u003cp\u003eThe authors have no relevant financial or non-financial interests to disclose.\u003c/p\u003e\n\u003cp\u003eAll behavioural trials were conducted under an animal care permit (#30019689) from Concordia University, Quebec, Canada.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe analysis code, scripts and supplementary figures and tables in this study are publicly archived on Zenodo at\u003c/p\u003e\n\u003cp\u003ehttps://doi.org/10.5281/zenodo.15734546\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eAlbins MA, Hixon MA (2008) Invasive Indo-Pacific lionfish Pterois volitans reduce recruitment of Atlantic coral-reef fishes. Marine Ecology Progress Series 367:233–238. https://doi.org/10.3354/meps07620.\u003c/li\u003e\n \u003cli\u003eBernardi G, Azzurro E, Bariche M et al. (2024) Invasion genomics of lionfish in the Mediterranean Sea. Ecology and Evolution 14:e11087. https://doi.org/10.1002/ece3.11087.\u003c/li\u003e\n \u003cli\u003eChristin S, Hervet É, Lecomte N (2019) Applications for deep learning in ecology. Methods in Ecology and Evolution 10:1632–1644. https://doi.org/10.1111/2041-210X.13256.\u003c/li\u003e\n \u003cli\u003eCôté IM, Smith NS (2018) The lionfish Pterois sp. invasion: Has the worst-case scenario come to pass? Journal of Fish Biology 92:660–689. https://doi.org/10.1111/jfb.13544.\u003c/li\u003e\n \u003cli\u003eChanthongdee K, Fuentealba Y, Wahlestedt T et al. (2024) Comprehensive ethological analysis of fear expression in rats using DeepLabCut and SimBA machine learning model. Frontiers in Behavioral Neuroscience 18:1440601. https://doi.org/10.3389/fnbeh.2024.1440601.\u003c/li\u003e\n \u003cli\u003eGreen SJ, Dilley ER, Benkwitt CE et al. (2019) Trait-mediated foraging drives patterns of selective predation by native and invasive coral-reef fishes. Ecosphere 10:e02752. https://doi.org/10.1002/ecs2.2752.\u003c/li\u003e\n \u003cli\u003eGreen SJ, Akins JL, Maljković A, Côté IM (2012) Invasive lionfish drive Atlantic coral-reef fish declines. PLoS One 7:e32596. https://doi.org/10.1371/journal.pone.0032596.\u003c/li\u003e\n \u003cli\u003eHe K, Zhang X, Ren S, Sun J (2016) Deep residual learning for image recognition. In: \u003cem\u003eProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)\u003c/em\u003e, pp 770–778. https://doi.org/10.1109/CVPR.2016.90.\u003c/li\u003e\n \u003cli\u003eHuang N, Lu G, Xu D (2016) A permutation importance-based feature selection method for short-term electricity load forecasting using random forest. Energies 9:767. https://doi.org/10.3390/en9100767.\u003c/li\u003e\n \u003cli\u003eHsu AI, Yttri EA (2021) B-SOiD: An open-source unsupervised algorithm for identification and fast prediction of behaviours. Nature Communications 12:5188. https://doi.org/10.1038/s41467-021-25420-x.\u003c/li\u003e\n \u003cli\u003eIPBES (2023) Summary for policymakers of the thematic assessment report on invasive alien species and their control. Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services, Bonn.\u003c/li\u003e\n \u003cli\u003eMathis A, Mamidanna P, Cury KM et al. (2018) DeepLabCut: Markerless pose estimation of user-defined body parts with deep learning. Nature Neuroscience 21:1281–1289. https://doi.org/10.1038/s41593-018-0209-y.\u003c/li\u003e\n \u003cli\u003eMorris JA Jr, Akins JL (2009) Feeding ecology of invasive lionfish (Pterois volitans) in the Bahamian archipelago. Environmental Biology of Fishes 86:389–398. https://doi.org/10.1007/s10641-009-9538-8.\u003c/li\u003e\n \u003cli\u003eNilsson SRO, Goodwin NL, Choong JJ et al. (2020) Simple Behavioural Analysis (SimBA): An open-source toolkit for computer classification of complex social behaviors in experimental animals. bioRxiv [pre-print]. https://doi.org/10.1101/2020.04.19.049452.\u003c/li\u003e\n \u003cli\u003ePereira TD, Aldarondo DE, Willmore L et al. (2019) Fast animal pose estimation using deep neural networks. Nature Methods 16:117–125. https://doi.org/10.1038/s41592-018-0234-5.\u003c/li\u003e\n \u003cli\u003eSoares MO, Pereira PHC, Feitosa CV et al. (2023) Lessons from the invasion front: Integration of research and management of the lionfish invasion in Brazil. Journal of Environmental Management 340:117954. https://doi.org/10.1016/j.jenvman.2023.117954.\u003c/li\u003e\n \u003cli\u003eSavitzky A, Golay MJE (1964) Smoothing and differentiation of data by simplified least-squares procedures. Analytical Chemistry 36:1627–1639. https://doi.org/10.1021/ac60214a047.\u003c/li\u003e\n \u003cli\u003eTang F, Ishwaran H (2017) Random forest missing-data algorithms. Statistical Analysis and Data Mining 10:363–377. https://doi.org/10.1002/sam.11348.\u003c/li\u003e\n \u003cli\u003eWeber RZ, Mulders G, Kaiser J et al. (2022) Deep learning-based behavioral profiling of rodent stroke recovery. BMC Biology 20:232. https://doi.org/10.1186/s12915-022-01434-9.\u003c/li\u003e\n \u003cli\u003eZhou Y, Liu Y, Li Z et al. (2021) Random forest model based on particle swarm optimization for prediction of surface roughness. Procedia Computer Science 184:393–401. https://doi.org/10.1016/j.procs.2021.03.050.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":true,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"marine-biology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"mabi","sideBox":"Learn more about [Marine Biology](https://www.springer.com/journal/227)","snPcode":"227","submissionUrl":"https://submission.nature.com/new-submission/227/3","title":"Marine Biology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Lionfish, Behaviour classification, Pose estimation, Machine learning DeepLabCut, SimBA","lastPublishedDoi":"10.21203/rs.3.rs-7153188/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7153188/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eModern neuroscience and ecology are increasingly adopting machine learning (ML) methods to automate the tracking and classification of animal behaviour. These techniques are particularly valuable for quantifying fitness‑related behaviours such as hunting in invasive predators. Here, we evaluate the effectiveness of two ML tools, DeepLabCut for pose estimation and SimBA for random‑forest behaviour classification, at distinguishing four behaviours in invasive lionfish (\u003cem\u003ePterois volitans\u003c/em\u003eand \u003cem\u003eP. miles\u003c/em\u003e): hovering, resting, swimming, and hunting, and benchmark the ML outputs against annotations from trained human observers. We also introduce a customized, user‑friendly feature‑extraction script tailored to lionfish. The script converts positional landmark coordinates extracted by DeepLabCut into a comprehensive set of kinematic metrics (e.g., body‑angle variance, fin‑beat frequency), essential for behaviour classification, as SimBA relies on these metrics rather than raw body‑part positions. A companion GitHub guide further clarifies which specific metrics are most informative under different behavioural scenarios. To our knowledge, this is the first study to explore how a wide‑angle camera lens influences the DeepLabCut–SimBA workflow. Behavioural trials, conducted in controlled aquarium settings, showed that the models classified high‑motion behaviours (hunting and swimming) with high precision and recall, likely owing to distinctive kinematic signatures. In contrast, low‑motion behaviours such as hovering and resting were harder to detect because of subtle movement cues, occasional suboptimal body orientation, and distortions introduced by the wide‑angle lens. While applicable to other mid-bodied fishes, lionfish were chosen due to their significant ecological impact, where quantifying behaviours like hunting can aid invasive species management and reef conservation.\u003c/p\u003e","manuscriptTitle":"Applying machine learning tools for automated behaviour classification in invasive lionfish and comparison with human observations","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-28 10:23:06","doi":"10.21203/rs.3.rs-7153188/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Acceptable after minor revision","date":"2025-09-16T04:10:38+00:00","index":"","fulltext":""},{"type":"reviewerAgreed","content":"","date":"2025-07-25T12:18:56+00:00","index":0,"fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-07-25T04:42:49+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-07-23T17:32:29+00:00","index":"","fulltext":""},{"type":"submitted","content":"Marine Biology","date":"2025-07-22T13:58:42+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"marine-biology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"mabi","sideBox":"Learn more about [Marine Biology](https://www.springer.com/journal/227)","snPcode":"227","submissionUrl":"https://submission.nature.com/new-submission/227/3","title":"Marine Biology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"a47defe1-2a1a-4ec5-8ccc-f08e37de65e0","owner":[],"postedDate":"July 28th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2026-02-23T16:00:22+00:00","versionOfRecord":{"articleIdentity":"rs-7153188","link":"https://doi.org/10.1007/s00227-025-04777-3","journal":{"identity":"marine-biology","isVorOnly":false,"title":"Marine Biology"},"publishedOn":"2026-02-16 15:57:02","publishedOnDateReadable":"February 16th, 2026"},"versionCreatedAt":"2025-07-28 10:23:06","video":"","vorDoi":"10.1007/s00227-025-04777-3","vorDoiUrl":"https://doi.org/10.1007/s00227-025-04777-3","workflowStages":[]},"version":"v1","identity":"rs-7153188","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7153188","identity":"rs-7153188","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}