OPTIMIZING MLOPs WITH CHAOS ENGINEERING AND SITE RELIABILITY ENGINEERING FOR ROBUST MACHINE LEARNING SYSTEMS
Fuente:
Zenodo
Enregistré dans:
| Auteur principal: | |
|---|---|
| Format: | Recurso digital |
| Publié: |
Zenodo
2025
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866901931274797056 |
|---|---|
| author | Pramod Begur Nagaraj |
| author_facet | Pramod Begur Nagaraj |
| contents | <p><em><span>One of the primary problems faced by the current ML systems is the ability to maximize performance and reliability in uncertain environments. Data sparsity is one of the main sources of trouble for the current methods as it limits the amount of input for the model. Another issue is that there are no predictive models available for dynamic environments and system failures are not handled well by the existing methods. The novel method put forth in this research work brings together Chaos Engineering, Decision Trees, and Site Reliability Engineering (SRE) principles to predict and enhance system optimization. The method is initiated with data gathering, where the focus is on the system performance data covering metrics such as response time based on the logs (which are also used as monitoring tools). For the collected data processing, the first step is to do missing values handling through mean imputation and then standardize it using Z-score normalization. Now comes Chaos Engineering where controlled failures are introduced to the system to simulate disruptions. With the chaotic nature of the disruptive failures, metrics such as recovery time, error rates, and latency under failure are collected. These features are then applied to train a Decision Tree model to predict the performance of the system and the likelihood of failure of the system. Evaluated model metrics include 99.51% Accuracy, 99.18% Precision, 98.97% Sensitivity & 99.26% F-Measure. Eventually, SRE principles are put to the test in monitoring and improving the system, by means of automated scaling and recovery capacity. This permits not only the resilience in the system to be strengthened but also the prediction metrics to be improved; thereby, a solid framework for systems management and optimization is presented. The research purposed to create and evaluate a hybrid MLOps optimization framework that unites the philosophies of Chaos Engineering and Site Reliability Engineering (SRE) for the purpose of breakthrough in prediction, experimentation, and avoidance of dynamic ML environments’ failures. The way is pioneering, thus yielding better systemic resilience and dependability by means of decision-oriented recovery and self-healing capability.</span></em></p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_18086415 |
| institution | Zenodo |
| language | |
| publishDate | 2025 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | OPTIMIZING MLOPs WITH CHAOS ENGINEERING AND SITE RELIABILITY ENGINEERING FOR ROBUST MACHINE LEARNING SYSTEMS Pramod Begur Nagaraj Chaos Engineering Decision Trees Site Reliability Engineering (SRE) System Optimization Performance Prediction Machine Learning <p><em><span>One of the primary problems faced by the current ML systems is the ability to maximize performance and reliability in uncertain environments. Data sparsity is one of the main sources of trouble for the current methods as it limits the amount of input for the model. Another issue is that there are no predictive models available for dynamic environments and system failures are not handled well by the existing methods. The novel method put forth in this research work brings together Chaos Engineering, Decision Trees, and Site Reliability Engineering (SRE) principles to predict and enhance system optimization. The method is initiated with data gathering, where the focus is on the system performance data covering metrics such as response time based on the logs (which are also used as monitoring tools). For the collected data processing, the first step is to do missing values handling through mean imputation and then standardize it using Z-score normalization. Now comes Chaos Engineering where controlled failures are introduced to the system to simulate disruptions. With the chaotic nature of the disruptive failures, metrics such as recovery time, error rates, and latency under failure are collected. These features are then applied to train a Decision Tree model to predict the performance of the system and the likelihood of failure of the system. Evaluated model metrics include 99.51% Accuracy, 99.18% Precision, 98.97% Sensitivity & 99.26% F-Measure. Eventually, SRE principles are put to the test in monitoring and improving the system, by means of automated scaling and recovery capacity. This permits not only the resilience in the system to be strengthened but also the prediction metrics to be improved; thereby, a solid framework for systems management and optimization is presented. The research purposed to create and evaluate a hybrid MLOps optimization framework that unites the philosophies of Chaos Engineering and Site Reliability Engineering (SRE) for the purpose of breakthrough in prediction, experimentation, and avoidance of dynamic ML environments’ failures. The way is pioneering, thus yielding better systemic resilience and dependability by means of decision-oriented recovery and self-healing capability.</span></em></p> |
| title | OPTIMIZING MLOPs WITH CHAOS ENGINEERING AND SITE RELIABILITY ENGINEERING FOR ROBUST MACHINE LEARNING SYSTEMS |
| topic | Chaos Engineering Decision Trees Site Reliability Engineering (SRE) System Optimization Performance Prediction Machine Learning |
| url | https://doi.org/10.5281/zenodo.18086415 |