cs.SEMar 24, 2026

Machine Learning Models for the Early Detection of Burnout in Software Engineering: a Systematic Literature Review

Authors: Tien Rahayu Tulili, Ayushi Rastogi, Andrea Capiluppi

Organizations: Bernoulli Institute for Mathematics, Computer Science, and Artificial Intelligence, University of Groningen, Nijenborgh 9, 9747AG, Groningen, Groningen, The Netherlands

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Burnout is an occupational syndrome that, like many other professions, affects the majority of software engineers. Past research studies showed important trends, including an increasing use of machine learning techniques to allow for an early detection of burnout. This paper is a systematic literature review (SLR) of the research papers that proposed machine learning (ML) approaches, and focused on detecting burnout in software developers and IT professionals. Our objective is to review the accuracy and precision of the proposed ML techniques, and to formulate recommendations for future researchers interested to replicate or extend those studies. From our SLR we observed that a majority of primary studies focuses on detecting emotions or utilise emotional dimensions to detect or predict the presence of burnout. We also performed a cross-sectional study to detect which ML approach shows a better performance at detecting emotions; and which dataset has more potential and expressivity to capture emotions. We believe that, by identifying which ML tools and datasets show a better performance at detecting emotions, and indirectly at identifying burnout, our paper can be a valuable asset to progress in this important research direction.

Figures & tables

Figure 1: Pipeline used in this SLR to obtain the relevant primary studies.
Inclusion Criteria
IC1The paper must be a peer-reviewed paper and published in conferences, workshops, or journals.
IC2The paper must be accessible online (i.e., PDF files available in the selected databases or through Google).
IC3The study must be published between the years 2000 and 2025.
IC4The study discussed or mentioned the early detection of burnout, or reported machine learning as its approach to predict burnout in software engineering.
Exclusion Criteria
EC1The non-English paper contains non-English data source used.
Table 1: Inclusion and Exclusion Criteria
No.QuestionsFocus
1Are the aims clearly stated?RQ1
2Does the paper clearly discuss the findings explaining the possibility of experiencing the risk of burnout or negative behaviour, or mention burnout or negative behaviour (e.g. negative emotion, feeling, stress or relationship)?RQ1
3Does the paper utilised the machine learning approach(s)?RQ1- RQ4
4Does the paper report the performance of the machine learning models?RQ3
5Does the paper clearly state the datasets employed?RQ4
Table 2: Assessment Criteria Form
CategoriesPaper IDMachine Learning Techniques
C1 Emotional Dimension as Burnout PredictorsMäntylä et al. (2016)Linear Regression Mäntylä et al. (2016) , Zero R Mäntylä et al. (2016)
C2 Burnout PredictionNath et al. (2021) , Dovleac et al. (2021)SVM Nath et al. (2021) , ANN Dovleac et al. (2021)
C3 Emotion. Dimension and Stress DetectionGirardi et al. (2020) ; Novielli et al. (2022) ; Nogueira et al. (2013) ; Girardi et al. (2021) ; Vrzakova et al. (2020) ; Fritz and Müller (2016) ; Nogueira et al. (2015) ; Naegelin et al. (2023) ; Epp et al. (2011) ; Srikanteswara et al. (2024) ; Dovleac et al. (2021) ; Gamage and Asanka (2022) ; Androutsou et al. (2023) ; Reddy et al. (2018) ; Booth et al. (2022) ; Padha and Sahoo (2022) ; Pepa et al. (2020) ; Alberdi et al. (2018) ; Vizer et al. (2009) ; Gachechiladze et al. (2017) ; Murgia et al. (2018) ; Imran (2024) ; Ballesteros et al. (2024) ; Bleyl and Buxton (2022) ; Maheshwarkar et al. (2021) ; Müller and Fritz (2015) ; Islam et al. (2019) ; Cabrera-Diego et al. (2020) ; Wagan and Li (2025) ; Yang et al. (2021) ; Singh et al. (2024) ; Soto et al. (2021) ; Anany et al. (2019) ; Klünder et al. (2020) ; Islam and Zibran (2018) ; Manikandan et al. (2024) ; Jayathilake et al. (2023) ; Carneiro et al. (2012) ; Rissler et al. (2020) , Manikandan et al. (2024) ; Geeth et al. (2024) ; Awan et al. (2023) ; Radevski et al. (2015) ; Kołakowska et al. (2013) ; Carneiro et al. (2012)AB Androutsou et al. (2023) ; Rissler et al. (2020) , ANN Srikanteswara et al. (2024) , BERT Bleyl and Buxton (2022) ; Maheshwarkar et al. (2021) ; Singh et al. (2024) ; Imran (2024) NN Nogueira et al. (2015) ; Kołakowska et al. (2013) , DT Nogueira et al. (2013) ; Islam et al. (2019) ; Fritz and Müller (2016) ; Nogueira et al. (2015) ; Epp et al. (2011) ; Srikanteswara et al. (2024) ; Jayathilake et al. (2023) ; Gamage and Asanka (2022) ; Reddy et al. (2018) ; Rissler et al. (2020) ; Müller and Fritz (2015) ; Kołakowska et al. (2013) , GBT Islam et al. (2019) , J48 Girardi et al. (2020) ; Gachechiladze et al. (2017) , KNN Girardi et al. (2020) ; Girardi et al. (2021) ; Srikanteswara et al. (2024) ; Androutsou et al. (2023) ; Reddy et al. (2018) ; Padha and Sahoo (2022) ; Rissler et al. (2020) ; Geeth et al. (2024) ; Kołakowska et al. (2013) ; Carneiro et al. (2012) , MLP Girardi et al. (2020) ; Islam et al. (2019) ; Srikanteswara et al. (2024) ; Booth et al. (2022) , NB Girardi et al. (2020) ; Gachechiladze et al. (2017) ; Fritz and Müller (2016) ; Islam et al. (2019) ; Srikanteswara et al. (2024) ; Gamage and Asanka (2022) ; Rissler et al. (2020) ; Gachechiladze et al. (2017) ; Maheshwarkar et al. (2021) ; Kołakowska et al. (2013) , RF Girardi et al. (2020) ; Soto et al. (2021) ; Novielli et al. (2022) ; Vrzakova et al. (2020) ; Girardi et al. (2021) ; Nogueira et al. (2013) ; Nogueira et al. (2015) ; Naegelin et al. (2023) ; Srikanteswara et al. (2024) ; Gamage and Asanka (2022) ; Maheshwarkar et al. (2021) ; Androutsou et al. (2023) ; Reddy et al. (2018) ; Booth et al. (2022) ; Anany et al. (2019) ; Rissler et al. (2020) , SLP Nogueira et al. (2013) ; Rissler et al. (2020) , SVM Girardi et al. (2020) ; Novielli et al. (2022) ; Nogueira et al. (2013) ; Gachechiladze et al. (2017) ; Fritz and Müller (2016) ; Nogueira et al. (2015) ; Naegelin et al. (2023) ; Srikanteswara et al. (2024) ; Androutsou et al. (2023) ; Padha and Sahoo (2022) ; Rissler et al. (2020) ; Gachechiladze et al. (2017) ; Murgia et al. (2018) ; Geeth et al. (2024) ; Islam et al. (2019) ; Kołakowska et al. (2013) ; Manikandan et al. (2024) , GBM Naegelin et al. (2023) ; Androutsou et al. (2023) , Logistic Regression Srikanteswara et al. (2024) ; Gamage and Asanka (2022) ; Reddy et al. (2018) ; Yang et al. (2021) , Linear Regression Srikanteswara et al. (2024) , LSTM Geeth et al. (2024) ; Maheshwarkar et al. (2021) ; Yang et al. (2021) , CNN Manikandan et al. (2024) ; Geeth et al. (2024) ; Ballesteros et al. (2024) ; Maheshwarkar et al. (2021) ; Wagan and Li (2025) , RNN Dovleac et al. (2021) ; Geeth et al. (2024) ; Manikandan et al. (2024) , XGBoost Gamage and Asanka (2022) ; Androutsou et al. (2023) , CatBoost Gamage and Asanka (2022) , RUSBoost Androutsou et al. (2023) , ZeroR Rissler et al. (2020) , Voting Classifier(SVM, NB, and RF) Klünder et al. (2020)
C4 Sentiment and EmotionsSilva et al. (2023) , Pletea et al. (2014) ; Ortu et al. (2016)Hierarchical Classifier Pletea et al. (2014) , NB Pletea et al. (2014) , RF Anany et al. (2019) , SentiStrength Silva et al. (2023) ; Ortu et al. (2016)
C5 Attrition predictionTrinkenreich et al. (2024) ; Ozakca et al. (2024) ; Garcia et al. (2013)AdaBoost Ozakca et al. (2024) , Bayesian Classfr. Garcia et al. (2013) DT Trinkenreich et al. (2024) ; Ozakca et al. (2024) , KNN Ozakca et al. (2024) , Logistic Regression Ozakca et al. (2024) , NB Ozakca et al. (2024) , RF Trinkenreich et al. (2024) ; Ozakca et al. (2024) , SVM Ozakca et al. (2024)
C6 Detecting Unhealthy Relationships as an Early Warning to Prevent BurnoutRaman et al. (2020) ; Cheriyan et al. (2021) ; Qiu et al. (2022) ; Sarker et al. (2023a) ; Ferreira et al. (2024) ; Sarker (2022) ; Mishra and Chatterjee (2024) ; Sarker et al. (2020) ; Bhat et al. (2021) ; Rahman et al. (2024)CART Ferreira et al. (2024) , ChatGPT Mishra and Chatterjee (2024) ; Rahman et al. (2024) CNN Sarker et al. (2020) ; Sarker (2022) ; Raman et al. (2020) , DNN Sarker et al. (2023b) , DT Sarker et al. (2023b) ; Sarker (2022) , GBT Sarker et al. (2023b) ; Sarker (2022) , KNN Ferreira et al. (2024) , LR Sarker et al. (2023b) , Lexicon-based Sarker et al. (2023a) , LR Ferreira et al. (2024) ; Sarker (2022) , NB Ferreira et al. (2024) , RF Cheriyan et al. (2021) ; Sarker et al. (2023b) ; Ferreira et al. (2024) ; Sarker (2022) ; Qiu et al. (2022) , SVM Raman et al. (2020) ; Sarker et al. (2020) ; Cheriyan et al. (2021) ; Sarker et al. (2023b) ; Ferreira et al. (2024) ; Sarker (2022) , BERT Sarker et al. (2020) ; Cheriyan et al. (2021) ; Sarker et al. (2023b) ; Sarker et al. (2023a) ; Sinha et al. (2016) ; Sarker (2022) ; Raman et al. (2020) ; Bhat et al. (2021) ; Rahman et al. (2024) , LSTM Sarker et al. (2023b) ; Sarker (2022)
Table 3: Table of analysed primary studies. Studies have employed more than one machine learning algorithm. Studies with bold fonts reported the performances in figures only or did not report the performances in their papers.
Figure 2: Number of papers grouped by purpose of study and input type.
 # of studiesstudies
Text-based  
 3Silva et al. (2023) ; Pletea et al. (2014) ; Ortu et al. (2016)
 24Mäntylä et al. (2016) ; Gachechiladze et al. (2017) ; Murgia et al. (2018) ; Islam and Zibran (2018) ; Imran (2024) ; Singh et al. (2024) ; Srikanteswara et al. (2024) ; Geeth et al. (2024) ; Gamage and Asanka (2022) ; Bleyl and Buxton (2022) ; Maheshwarkar et al. (2021) ; Islam et al. (2019) ; Reddy et al. (2018) ; Cabrera-Diego et al. (2020) ; Wagan and Li (2025) ; Muñoz and Iglesias (2022) ; Klünder et al. (2020) ; Nath et al. (2021) ; Vizer et al. (2009) , Girardi et al. (2020) ; Girardi et al. (2021) ; Müller and Fritz (2015) ; Androutsou et al. (2023) ; Yang et al. (2021)
 11Raman et al. (2020) ; Cheriyan et al. (2021) ; Qiu et al. (2022) ; Sarker et al. (2023b) ; Sarker et al. (2023a) ; Ferreira et al. (2024) ; Sarker (2022) ; Mishra and Chatterjee (2024) ; Sarker et al. (2020) ; Bhat et al. (2021) ; Rahman et al. (2024)
 3Trinkenreich et al. (2024) ; Ozakca et al. (2024) ; Garcia et al. (2013)
Sensor-based  
Table 4: The list of studies grouped by their types of input and classification. All studies that employ sensors, movement, utterances, and facial expressions as their input are studies investigating emotions and stress detection. Bold studies use multiple input types, while regular studies use just one.
Figure 3: Model Performances of emotion and stress detection with machine learning techniques categorised based on their most generalised characteristics. Bayesian (e.g. Naïve Bayes), Lexicon based (e.g. Deva, Tensistrength, EmoTxt), Ensemble method (e.g. Adaptive Boast, XGBoost, CatBoost, Boosting), Instance-based algorithm (e.g. KNN), Linear Model (e.g. Logistic Regression), NN-based algorithms (e.g. MLP, SLP, Sentimoji, CNN), Ruled-based Algorithm (e.g. ZeroR), Tree-based Algorithms (e.g. DT, J48, RF), and Transformer-based Algorithms (e.g., GoEmotion, BERT, Albert, Roberta, Bert, Codebert, Graphcodebert)
Figure 4: Model Performances of Emotion and Stress Detection with sensor and movement data as the inputs and machine learning techniques categorised based on their most generalised characteristics. Bayesian (e.g. Naïve Bayes, Bayes Net), Ensemble method (e.g., AdaBoast, Light Gradient Boasting Method), Instance-based algorithm (e.g. KNN, K-star, IBk), Linear Model (e.g., Logistic Regression), NN-based algorithms (e.g. CNN, RestNet), Ruled-based Algorithm (e.g. ZeroR), Tree-based Algorithms (e.g., DT, J48, RF, C.45)
Figure 5: Model Performances of attrition detection with machine learning techniques categorised based on their most generalised characteristics. Bayesian (e.g. Naïve Bayes), Instance-based algorithm (e.g. KNN), Linear Model (e.g., Logistic Regression), NN-based algorithms (e.g. MLP), Tree-based Algorithms (e.g., DT, RF), Ensemble (e.g., Adaptive Boost), and Kernel-based algorithms (e.g., SVM, SVC)
Figure 6: Model Performances of toxicity detection with machine learning techniques categorised based on their most generalised characteristics. Bayesian (e.g. Naïve Bayes), Instance-based algorithm (e.g. KNN), Linear Model (e.g., Logistic Regression), NN-based algorithms (e.g. Deep Pyramid CNN, Strudel, DPCNN, DCRNN, LSTM, BiLSTM, and GRU), Tree-based Algorithms (e.g., DT, GBT, RF, CART), and Transformed-based model (e.g., BERT, RoBERTa, DistilBERT, ALBERT, XLNet, ChatGPT, and GPT4.0)
 PrecisionRecallF-scoreAccuracy
Emotion and Stress detection (using text)Transformer-based model, Kernel-basedTransformer-based model, Kernel-based modelKernel-based modelBayesian
Emotion and Stress detection (using sensors and movement)BayesianBayesianInstance-based modelEnsemble
Attrition PredictionTree-based modelTree-based modelTree-based, EnsembleTree-based
Toxicity detectionNN-based model, Transformer-based modelTransformer-based modelNN-based modelNN-based model, Tree-based model
Table 5: Summary of performance measurements - RQ3
Figure 7: Model Performances by datasets
 PrecisionRecallF-scoreAccuracy
BugzillaMMM 
Combination   M
Code Review (CR)HHHH
GerritHhhh
GithubhhML
GitterHHHH
Table 6: Summary of performance measurements - RQ4. H, h, M, and L represent the performance value (median) with the criteria respectively: more than 0.8; between 0.7 and 0.8; between 0.5 and 0.7; below 0.5
Figure 8: Trends of classification during the period
Figure 9: A framework of machine-learning based early detection of burnout
More in this topic

Software Engineering

Software engineering studies how to design, build, test, deploy, and maintain reliable software systems and the processes behind them.

See all 270 papers

Explore similar work

CardsList
Sep 16, 2026

BurnRiSc: Toward Non-Invasive Burnout Screening in Open Source from Public Repository Signals

Burnout is a chronic occupational syndrome, and open source is close to a worst case for it: maintainers absorb unbounded demand with no manager to reallocate work and no organization to notice decline. The cost is not only personal. Burnout precedes withdrawal, and in projects sustained by a handful of maintainers, one departure can break infrastructure that thousands of downstream systems depend on. Yet the field has no way to see it coming: self-report inventories, the only existing measure, miss exactly the contributors most in need of detection and cannot be applied retroactively, so the field cannot even ask how common burnout is or what helps. We present BurnRiSc, a framework that operationalizes the Oldenburg Burnout Inventory's two dimensions, exhaustion and disengagement, as 14 behavioral and linguistic signals computed from GitHub activity and scored against each contributor's own history. The signals aggregate into two weighted dimension scores, with weights learned from labeled cases, and average into a monthly Burnout Risk Score (BRS). In a preliminary evaluation across 68 contributors in ten repositories (ten disclosed burnout cases, twelve comparable-volume collapses, and 46 comparison contributors), sustained BRS elevation precedes 6 of 10 disclosures by 6-15 months, 8 of 10 when adding peak BRS as a second criterion, and 10 of 10 over any prior time frame. We thus present BurnRiSc as evidence that burnout is screenable from public data.

Timofey Sanko, Yuan Tian, Mariam Guizani

Queen’s University Kingston, Ontario, Canada

Software EngineeringOpen-Source Software DevelopmentMental Health
AskPDF
Apr 15, 2026

Sentiment analysis for software engineering: How far can zero-shot learning (ZSL) go?

Sentiment analysis in software engineering focuses on understanding emotions expressed in software artifacts. Previous research highlighted the limitations of applying general off-the-shelf sentiment analysis tools within the software engineering domain and indicated the need for specialized tools tailored to various software engineering contexts. The development of such tools heavily relies on supervised machine learning techniques that necessitate annotated datasets. Acquiring such datasets is a substantial challenge, as it requires domain-specific expertise and significant effort. Objective: This study explores the potential of ZSL to address the scarcity of annotated datasets in sentiment analysis within software engineering Method:} We conducted an empirical experiment to evaluate the performance of various ZSL techniques, including embedding-based, NLI-based, TARS-based, and generative-based ZSL techniques. We assessed the performance of these techniques under different labels setups to examine the impact of label configurations. Additionally, we compared the results of the ZSL techniques with state-of-the-art fine-tuned transformer-based models. Finally, we performed an error analysis to identify the primary causes of misclassifications. Results: Our findings demonstrate that ZSL techniques, particularly those combining expert-curated labels with embedding-based or generative-based models, can achieve macro-F1 scores comparable to fine-tuned transformer-based models. The error analysis revealed that subjectivity in annotation and polar facts are the main contributors to ZSL misclassifications. Conclusion: This study demonstrates the potential of ZSL for sentiment analysis in software engineering. ZSL can provide a solution to the challenge of annotated dataset scarcity by reducing reliance on annotated dataset.

Reem Alfayez, Manal Binkhonain

Department of Software Engineering, College of Computer and Information Sciences, King Saud University, P.O. Box 51178, Riyadh 11543, Saudi Arabia

Sentiment AnalysisZero-Shot Text ClassificationZero-Shot Learning
AskPDF
Aug 10, 2026

Detecting Soft Skills in ML Engineering Roles CVs

Soft skills shape collaboration among ML engineers, data scientists, and software engineers building ML-enabled systems, yet what we know about them comes almost entirely from the demand side. Job advertisements, surveys, and hiring manager interviews capture what employers ask for. How candidates themselves articulate these competencies has not been studied, and existing CV-mining work is both keyword-based, so it cannot see skills conveyed through narrative, and descriptive, reporting frequency rankings without testing whether group differences exceed sampling variation. We close both gaps. Using a balanced corpus of 300 curated CVs spanning the three roles, we extract explicitly listed and implicitly narrated soft skills with an LLM-based pipeline validated against a human-annotated ground truth, a distinction that existing extractors were not designed to make. We then convert the demand-side literature's claims into 13 falsifiable hypotheses about role signatures, seniority progression, and disclosure style, and test them with effect sizes under family-wise error control, so that candidate-side data can corroborate or contradict the demand-side account rather than merely illustrate it. Eleven hypotheses are supported, one partially, and one refuted. Candidates disclose soft skills through narrative rather than keyword lists by roughly three to one, and most so for the competencies employers value most: leadership, coordination, and mentoring (88-96% narrative). Seniority nearly triples the odds of articulating leadership. That competency, assumed universal in prior work, is articulated by software engineers at half the rate of their peers. Technical candidates do articulate soft skills, but a keyword-based screening systematically misses them.

Aidin Azamnouri, Nouran Ayad, Justus Bogner +1

Chair of Software Engineering, TUM School of Computation, Information and Technology, Technical University of Munich, Heilbronn, Germany · Software and Sustainability (S2) Group, Vrije Universiteit Amsterdam, Amsterdam, The Netherlands

Software Engineering
AskPDF