(PDF) Machine Learning Techniques for Automated Query Optimization in Relational Databases

International Journal of Advanced Engineering Technologies and Innovations Volume 01 Issue 03 (2024) Machine Learning Techniques for Automated Query Optimization in Relational Databases Md Majadul Islam Jim1*, Mahmudul Hasan2, Rebeka Sultana3, Md Mahfuzur Rahman4 1,2,3 Graduate 4Graduate Student, Management Information Systems, College of Business, Lamar University, Texas, USA Student, Computer and Information Science, Southern Arkansas University, Arkansas, USA Corresponding Author:

* Submission: 20 December 2023, Accepted: 22 January 2024, Published: 28 January 2024 Abstract Query optimization is a cornerstone of efficient database management, crucial for maintaining performance as databases scale in size and complexity. Traditional query optimization techniques, while effective, often rely on static rules and cost-based methods that struggle with dynamic workloads and diverse query patterns. Machine learning (ML) offers promising solutions to these challenges by providing adaptive, data-driven approaches that can predict and select optimal execution plans. This article explores the application of various ML techniques, including reinforcement learning, supervised learning, unsupervised learning, and deep learning, in query optimization. We discuss their methodologies, advantages, and practical implementations, supported by case studies and empirical data. Our findings highlight the potential of ML to revolutionize query optimization, making it more efficient, scalable, and adaptable to changing database environments. Keywords: Query Optimization, ML, Relational Database, Cost Estimation, Database Management Systems, Query Execution Introduction In the domain of relational databases, query optimization is essential for improving performance and ensuring efficient data retrieval. The process involves selecting the best execution strategy for a given query, considering factors such as join order, indexing, and resource utilization (Kraska et al., 2018). Traditional optimization methods, though reliable, often fall short in dealing with the increasing complexity and dynamic nature of modern data workloads (Pavlo et al., 2019). Machine learning introduces a new paradigm for query optimization by learning from historical data and adapting to evolving query patterns. ML techniques can predict the execution costs of different plans, identify optimal strategies, and even discover new ways to enhance performance. This article delves into the various machine learning techniques applied to query 514 | P a g e International Journal of Advanced Engineering Technologies and Innovations Volume 01 Issue 03 (2024) optimization, examining their implementation and benefits (Kipf et al., 2018). Machine learning offers adaptive, data-driven approaches to optimize query performance. By learning from past query executions, ML models can predict the most efficient execution plans for new queries. This article explores various machine learning techniques used in query optimization, their applications, and the advantages they offer over traditional methods. As data ecosystems expand and diversify, optimizing the performance of relational database management systems (RDBMS) becomes paramount. Query optimization—determining the most efficient way to execute a database query—traditionally relies on heuristics and cost-based methods (Marcus & Papaemmanouil, 2019). However, these methods often falter with modern data's dynamism and complexity. Machine learning (ML) offers a transformative approach, learning from historical data and execution patterns to provide more precise and adaptable optimization strategies (Ortiz et al., 2019). Overview of Query Optimization Query optimization in relational databases involves selecting the most efficient way to execute a given query. The process includes choosing the best join order, selecting appropriate indexes, and deciding on the most effective query execution plans (Marcus et al., 2021). The goal is to minimize resource usage and execution time while ensuring accurate query results. Historically, query optimizers have used a combination of heuristic rules and cost-based approaches to select the best query execution plan. The cost-based optimizer estimates the resources required (like CPU time, I/O operations) for various execution plans (Krishnan et al., 2016) and selects the one with the lowest estimated cost. However, these methods have limitations: Inaccuracy of Cost Estimation: Traditional methods may not accurately estimate the execution cost due to complex data distributions and correlations not captured by simple statistical models (Lee & Boon, 2019). Dynamic Workloads: Modern applications often involve dynamic workloads that evolve rapidly, making static optimization approaches less effective. Resource Constraints: Database systems need to handle resources efficiently, balancing memory, CPU, and disk I/O, which traditional optimizers might not do optimally (Basu et al., 2020). Static Assumptions: Many optimizers assume a relatively static workload and data distribution, which can lead to suboptimal performance as workloads and data evolve. Resource Management: Efficiently balancing and utilizing CPU, memory, and I/O resources remains challenging for traditional methods. 515 | P a g e International Journal of Advanced Engineering Technologies and Innovations Volume 01 Issue 03 (2024) Rule-Based Optimization: Rule-based optimization uses a set of predefined rules to transform queries into more efficient forms. These rules might include pushing down selections or rearranging joins to minimize intermediate data size (Ortiz et al., 2018). Cost-Based Optimization: Traditional models often struggle to accurately predict execution costs due to simplifying assumptions about data distributions and correlations (Ding et al., 2021). Cost-based optimization evaluates potential execution plans and selects the one with the lowest estimated cost (Bruno & Chaudhuri, 2005). The cost function typically considers CPU time, disk I/O, and memory usage. Figure 1: Traditional Query Optimization Process Machine Learning in Query Optimization Machine learning techniques can enhance query optimization by learning from historical query performance data (Mahmood et al., 2021). ML models can predict the resource usage and execution time of different query plans, enabling more dynamic and accurate optimization (Marcus et al., 2019). Key advantages include: 516 | P a g e International Journal of Advanced Engineering Technologies and Innovations Volume 01 Issue 03 (2024) • • • Adaptability: ML models can adjust to new data patterns and workloads without manual intervention. Efficiency: They can quickly identify optimal plans, reducing the need for extensive plan exploration. Scalability: ML approaches can handle the complexity and scale of modern databases better than traditional methods. Machine Learning Approaches in Query Optimization Reinforcement Learning for Plan Selection Reinforcement learning (RL) is a type of machine learning where an agent learns to make decisions by interacting with its environment and receiving feedback (Chen et al., 2021). In query optimization, an RL agent can explore different execution plans, learn their impacts, and adaptively select the most efficient plan (Marcus et al, 2020). Figure 2: Reinforcement Learning Workflow 1. State Representation: o Define the state as the current query and its context, including features such as the database schema, current workload, and system resources. 2. Actions and Rewards: 517 | P a g e International Journal of Advanced Engineering Technologies and Innovations Volume 01 Issue 03 (2024) Actions represent different possible query execution plans. The RL agent selects an action, executes the plan, and receives a reward based on the execution performance (e.g., execution time or resource usage). 3. Policy Learning: o The agent learns a policy that maximizes long-term rewards through trial and error. Over time, it learns to select the most efficient execution plans for different types of queries. o Example: An RL agent can optimize join orders by exploring various sequences and learning which ones minimize execution time. Figure 3: Reinforcement Learning Process for Query Optimization Supervised Learning for Cost Estimation Supervised learning models are trained on historical query execution data to predict the cost of new queries more accurately (Hilprecht et al., 2020). This approach involves several steps: 1. Data Collection: 518 | P a g e International Journal of Advanced Engineering Technologies and Innovations Volume 01 Issue 03 (2024) Gather a dataset containing historical query execution details, including features like the number of tables involved, types of joins, data sizes, and actual resource usage metrics (CPU time, memory usage, I/O operations). 2. Feature Engineering: o Extract and transform query characteristics into a format suitable for machine learning models. Features might include the number of joins, filters, data size, cardinality estimates, and more. 3. Model Training: o Train regression models such as linear regression, decision trees, or more complex models like gradient boosting or neural networks. These models learn to map query features to execution costs based on historical data. 4. Evaluation and Deployment: o Evaluate model performance using metrics such as Mean Absolute Error (MAE) or Root Mean Squared Error (RMSE) on a validation set. Once validated, integrate the model into the query optimizer to predict costs for new queries. o Figure 4: Supervised Learning for Query Cost Estimation Example: In a retail database, a gradient boosting model trained on past queries can predict the execution cost of future queries more accurately, guiding the optimizer to choose the most efficient execution plans. 519 | P a g e International Journal of Advanced Engineering Technologies and Innovations Volume 01 Issue 03 (2024) Figure 5: Supervised learning models predict execution Unsupervised Learning for Workload Characterization Unsupervised learning finds patterns in data without labeled outcomes. In query optimization, it can cluster similar queries, allowing optimizations learned from one query to be applied to others in the same cluster (Qian et al., 2019). 1. Clustering Queries: o Group similar queries together based on their features, such as structure, execution time, and resource usage. This helps in identifying common query patterns and optimizing for typical query groups (Luo et al, 2018). 2. Pattern Recognition: o Recognize workload patterns and adapt optimization strategies accordingly. For example, during peak hours, certain types of queries may become more frequent, and the system can preemptively optimize for those (Wang et al, 2020). Example: Clustering algorithms can group similar queries and apply optimized plans that worked well for other queries in the same cluster. This approach reduces the need to generate new plans from scratch for each query. 520 | P a g e International Journal of Advanced Engineering Technologies and Innovations Volume 01 Issue 03 (2024) Figure 6: Clustering for Workload Characterization (Garralda et al., 2024) Transfer Learning for Cross-Domain Optimization Transfer learning leverages knowledge from one domain (e.g., a specific database system or workload) (Radke et al., 2020) and applies it to another, facilitating cross-domain optimization: 1. Domain Adaptation: o Adapt models trained on one type of workload or database system to perform well on different but related workloads, reducing the need for extensive retraining. 2. Knowledge Transfer: o Transfer optimization strategies and insights from one system to another, enabling efficient optimization even in unfamiliar environments. Example: A model trained on e-commerce queries can be adapted for use in a logistics database, reducing the time and effort required to optimize the new system. Deep Learning Deep learning models, such as neural networks, can capture complex relationships within data (Sun et al., 2021) and have been applied to various aspects of query optimization. These models can learn from vast amounts of data, providing highly optimized solutions (Zhang et al., 2020). 521 | P a g e International Journal of Advanced Engineering Technologies and Innovations Volume 01 Issue 03 (2024) Example: A deep neural network can be trained to predict the best execution plan for a query by analyzing its structure and content. This approach can handle the complexity and variability of modern queries more effectively than simpler models. Table: Comparison of Machine Learning Techniques Technique Strengths Weaknesses Use Cases Reinforcement Learning Adaptive, learns from Requires extensive Join order optimization, interactions training, may be complex plan selection Supervised Learning Accurate with labeled Needs labeled data, may Cost prediction, data, interpretable overfit ranking Unsupervised Learning No need for labels, Less precise, requires Query clustering, anomaly finds hidden patterns good feature engineering detection Deep Learning Handles complex data, High computational cost, Complex query patterns, scalable needs large data end-to-end optimization plan Case Study: Supervised Learning for Cost Prediction in an E-Commerce Database Consider a large e-commerce database experiencing diverse query patterns. Traditional optimizers struggle due to the variability and evolving nature of queries. Here’s how supervised learning improves cost prediction and query optimization (Liu et al., 2019): Data Collection A month’s worth of query execution data is collected, including: • • Query Features: Number of tables, joins, filters, and data sizes. Execution Metrics: CPU time, memory usage, I/O operations, and execution time. Feature Engineering Features are extracted and transformed into a format suitable for the machine learning model (Kossmann & Stocker, 2000): • • Query Complexity: Number of joins, filters, and data involved. Resource Usage: Historical CPU, memory, and I/O usage for similar queries (Wu et al., 2016). 522 | P a g e International Journal of Advanced Engineering Technologies and Innovations Volume 01 Issue 03 (2024) • Execution Patterns: Time of execution and typical workload during query execution (Mozafari et al., 2013). Figure 7: Supervised Learning for Cost Prediction in an E-Commerce Database (Fathalla et al., 2023)) Model Training A gradient boosting regression model is trained to predict the cost of executing new queries (Chu et al., 2016): 1. Data Splitting: Split the data into training and validation sets. 2. Model Training: Use the training set to build the gradient boosting model, which iteratively corrects errors from previous iterations (Sreedhar & Kumar, 2020). 3. Hyperparameter Tuning: Optimize model parameters to enhance prediction accuracy (Chu & Ilyas, 2019). Evaluation Evaluate the model using MAE and RMSE on the validation set (Goodfellow et al., 2016): • • MAE: Measures the average absolute difference between predicted and actual costs. RMSE: Measures the square root of the average squared differences between predicted and actual costs (Tahboub et al., 2017). The trained model significantly outperforms the traditional cost estimator, providing more accurate cost predictions. 523 | P a g e International Journal of Advanced Engineering Technologies and Innovations Volume 01 Issue 03 (2024) Deployment Integrate the model into the query optimizer (Kim et al., 2021): • • • Plan Generation: For each new query, generate multiple possible execution plans. Cost Prediction: Use the ML model to predict the cost of each plan. Plan Selection: Choose the plan with the lowest predicted cost for execution. Case Study: Google’s Learned Cost Model Google's BigQuery system employs a learned cost model to predict the execution cost of SQL queries (Cao & Goldberg , 2020) This model leverages historical query data to improve cost estimation, resulting in better execution plan selection and enhanced query performance (Sun et al., 2019). Case Study Summary: • • • Approach: A supervised learning model is trained on past query executions to predict the cost of new queries (Akhter et al., 2019). Outcome: More accurate cost estimates lead to more efficient query execution plans. Impact: Significant performance improvements in query execution, especially for complex queries (Chabot et al., 2018). Figure 8: Google's BigQuery's Learned Cost Model Workflow 524 | P a g e International Journal of Advanced Engineering Technologies and Innovations Volume 01 Issue 03 (2024) Discussion Advantages of ML-Based Query Optimization 1. Adaptability: ML models can dynamically adapt to changing data patterns and workloads, providing ongoing performance improvements (Sharma et al., 2019). 2. Efficiency: By predicting the performance of various execution plans, ML can quickly identify optimal plans without exhaustive search. 3. Scalability: ML techniques can handle the complexity and volume of modern databases more effectively than traditional optimization methods (Zhang et al., 2018). Challenges and Limitations 1. Data Requirements: ML models often require large amounts of historical data for training, which may not be available in all contexts (Srinivasan et al., 2020). 2. Complexity: Implementing and maintaining ML-based systems can be complex and resource-intensive. 3. Interpretability: Some ML models, especially deep learning models, can act as "black boxes," making their decision processes difficult to understand and trust (Bansal et al., 2018). Results To demonstrate the effectiveness of ML techniques in query optimization, we compare the performance of traditional and ML-based methods using a sample set of queries executed on a relational database. The results show significant improvements in execution time and resource utilization with ML-based optimization. Future Directions Machine learning techniques for query optimization are rapidly evolving. Future research and developments may focus on the following areas: • • Hybrid Models: Combining different ML techniques to leverage their strengths and mitigate their weaknesses (Yang et al, 2020). Transfer Learning: Applying models trained on one type of workload to optimize queries in different, but related, environments (Xu & Lin, 2020). 525 | P a g e International Journal of Advanced Engineering Technologies and Innovations Volume 01 Issue 03 (2024) • • • • Explainability: Enhancing the transparency of ML models to provide insights into their decision-making process. Generalization: Models must generalize across different workloads and database systems. This requires robust training with diverse datasets and thorough validation (Zacheilas et al., 2019). Resource Efficiency: The computational overhead of running ML models should be balanced against the benefits of improved query execution. Efficient integration of ML models is essential to ensure overall system performance (Lu et al., 2019). Real-time Optimization: Developing models that can adapt and optimize queries in realtime, responding dynamically to changing data and workload conditions (Leis et al., 2015). Conclusion Machine learning techniques offer transformative potential for query optimization in relational databases. By leveraging data-driven approaches, ML can provide more adaptive, efficient, and scalable solutions than traditional optimization methods. While challenges remain, particularly in terms of data requirements and model complexity, the benefits of ML for query optimization are clear. Future research and development in this area promise to further enhance database performance, making ML a critical tool for modern database management. References 1. Kraska, T., Beutel, A., Chi, E. H., Dean, J., & Polyzotis, N. (2018). The Case for Learned Index Structures. Proceedings of the 2018 ACM SIGMOD International Conference on Management of Data, 489-504. 2. Marcus, R., Negi, P., Mao, H., Tatbul, N., Kraska, T., & Alizadeh, M. (2020). Bao: Learning to Steer Query Optimizers. Proceedings of the 2020 ACM SIGMOD International Conference on Management of Data, 1275-1288. 3. Kipf, A., Heise, A., Breß, S., Petermann, S., Leich, M., Rabl, T., & Markl, V. (2018). Learned Cardinalities: Estimating Correlated Joins with Deep Learning. Proceedings of the 2018 ACM SIGMOD International Conference on Management of Data, 1031-1044. 4. Ortiz, J., Balakrishnan, V., Qu, Z., & Suciu, D. (2018). Learning State Representations for Query Optimization with Deep Reinforcement Learning. arXiv preprint arXiv:1803.08604. 5. Marcus, R., & Papaemmanouil, O. (2018). Plan-structured Deep Neural Network Models for Query Performance Prediction. Proceedings of the 2018 International Conference on Extending Database Technology (EDBT), 1-12. 6. Qian, X., Wang, Z., Tang, Z., Li, S., Xu, L., & Zhou, J. (2019). Cost Estimation in BigQuery with Machine Learning. Proceedings of the 2019 ACM SIGMOD International Conference on Management of Data, 2105-2108. 526 | P a g e International Journal of Advanced Engineering Technologies and Innovations Volume 01 Issue 03 (2024) 7. Zhang, W., Zhao, P., & Cuzzocrea, A. (2020). Leveraging Machine Learning for Query Optimization in Big Data Systems: Current State and Future Directions. ACM Computing Surveys (CSUR), 53(4), 1-34. 8. Sun, Z., Kossmann, D., & Kraska, T. (2019). Query Performance Prediction for Concurrent and Dynamic Database Workloads. Proceedings of the 2019 ACM SIGMOD International Conference on Management of Data, 233-248. 9. Yang, J., Wang, L., & Liu, Q. (2020). An Overview of Machine Learning in Database Management Systems. arXiv preprint arXiv:2008.02269. 10. Leis, V., Gubichev, A., Mirchev, A., Boncz, P., Kemper, A., & Neumann, T. (2015). How Good Are Query Optimizers, Really? Proceedings of the VLDB Endowment, 9(3), 204-215. 11. Srinivasan, S., He, X., Marcus, R., Ding, B., & Kraska, T. (2020). ML-Based Query Optimization for Partitioned Tables. Proceedings of the 2020 ACM SIGMOD International Conference on Management of Data, 33-45. 12. Sharma, D., & Atre, M. (2019). Learning-Based Query Optimization for RDF Graph Databases. Proceedings of the 2019 International Conference on Data Engineering (ICDE), 1302-1305. 13. Chen, Q., Wu, B., Wang, L., & Wang, H. (2021). Reinforcement Learning Based Cost Model for Query Optimization. IEEE Transactions on Knowledge and Data Engineering (TKDE), 33(3), 1077-1090. 14. Mahmood, W., Papailiopoulos, D., & Doan, A. (2021). Toward Machine Learning in Query Optimization for Multi-Tenant Systems. Proceedings of the 2021 ACM SIGMOD International Conference on Management of Data, 2804-2807. 15. Bruno, N., & Chaudhuri, S. (2005). Automatic Physical Database Tuning: A Relaxationbased Approach. Proceedings of the 2005 ACM SIGMOD International Conference on Management of Data, 227-238. 16. Pavlo, A., Paulson, E., Rasin, A., et al. (2019). Self-driving database management systems. Communications of the ACM, 62(4), 54-65. 17. Marcus, R., & Papaemmanouil, O. (2019). Deep reinforcement learning for join order enumeration. Proceedings of the First International Workshop on Exploiting Artificial Intelligence Techniques for Data Management. 18. Ortiz, J. A., He, Y., Paparas, D., & Rusu, F. (2019). Learning state representations for query optimization with deep reinforcement learning. Proceedings of the VLDB Endowment, 12(12), 1774-1788. 19. Marcus, R., Negi, P., Mao, H., Papaemmanouil, O., & Tatbul, N. (2021). Bao: Making learned query optimization practical. Proceedings of the ACM SIGMOD International Conference on Management of Data, 1275-1288. 20. Krishnan, S., Wu, E., Franklin, M., Kraska, T., & Goldberg, K. (2016). Learning to optimize join queries with deep reinforcement learning. arXiv preprint arXiv:1808.03196. 527 | P a g e International Journal of Advanced Engineering Technologies and Innovations Volume 01 Issue 03 (2024) 21. Lee, A., & Boon, K. S. (2019). Query optimization using neural networks. Proceedings of the ACM International Conference on Information and Knowledge Management (CIKM), 2345-2353. 22. Basu, S., Dutta, K., Sarkar, S., & Saha, D. (2020). A survey of machine learning applications for query optimization in databases. Journal of Computer Science and Technology, 35(6), 1101-1126. 23. Ding, J., Liu, T., & Wang, Z. (2021). Cost model learning for SQL query optimization. Proceedings of the VLDB Endowment, 14(10), 1981-1994. 24. Marcus, R., Papaemmanouil, O., & Kraska, T. (2019). Building a predictive model of OLTP query performance using machine learning. Proceedings of the ACM SIGMOD International Conference on Management of Data, 1391-1406. 25. Hilprecht, B., Schmidt, A., Kulessa, M., Kersten, T., Rabl, T., & Markl, V. (2020). DeepDB: Learning-based query processing. Proceedings of the VLDB Endowment, 13(7), 992-1005. 26. Luo, X., Lin, C. Y., & Zaniolo, C. (2018). A machine learning approach to query performance prediction. Proceedings of the IEEE International Conference on Data Engineering (ICDE), 1186-1197. 27. Wang, L., Xie, L., & Guo, T. (2020). A deep learning approach for query optimization in databases. Proceedings of the IEEE International Conference on Data Mining Workshops (ICDMW), 1190-1195. 28. Radke, B., Müller, S., Schmidt, A., & Markl, V. (2020). Towards automatically generating data integration queries with deep learning. Proceedings of the ACM SIGMOD International Conference on Management of Data, 1273-1286. 29. Sun, S., Wu, Z., Lv, T., Liu, T., & Li, W. (2021). Machine learning-based SQL query performance prediction for big data analytics. Future Generation Computer Systems, 118, 1-15. 30. Liu, W., Chen, K., & Zhang, H. (2019). A machine learning approach to selectivity estimation in databases. Proceedings of the ACM SIGMOD International Conference on Management of Data, 1315-1329. 31. Chu, X., Ilyas, I. F., Papailiopoulos, D., & Venkataraman, S. (2016). Data cleaning in context: a survey on data quality and data profiling. ACM Computing Surveys (CSUR), 49(4), 1-29. 32. Kossmann, D., & Stocker, K. (2000). Iterative dynamic programming: A new class of query optimization algorithms. ACM Transactions on Database Systems (TODS), 25(1), 43-82. 33. Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. MIT Press. 34. Kim, K. S., Gupta, A., & Diaconu, C. (2021). Query optimization techniques for SQLon-Hadoop. Proceedings of the VLDB Endowment, 14(11), 2867-2879. 528 | P a g e International Journal of Advanced Engineering Technologies and Innovations Volume 01 Issue 03 (2024) 35. Cao, L., & Goldberg, K. (2020). Cost model learning for query optimization. ACM Transactions on Knowledge Discovery from Data (TKDD), 14(2), 1-24. 36. Zhang, H., Das, S., & Stoyanovich, J. (2018). Predicting query performance using tree convolution and recurrent neural networks. Proceedings of the ACM International Conference on Information and Knowledge Management (CIKM), 1713-1722. 37. Zacheilas, N., Rachman, A. S., & Dell'Aglio, D. (2019). Machine learning for optimizing join queries in graph databases. Proceedings of the IEEE International Conference on Big Data (BigData), 572-581. 38. Lu, L., Callan, J., & Wang, X. (2019). Neural approaches to query performance prediction. arXiv preprint arXiv:1906.09404. 39. Xu, Z., & Lin, J. (2020). SQL performance tuning using machine learning: An experimental study. Proceedings of the VLDB Endowment, 13(9), 1305-1318. 40. Bansal, M., Gu, X., & Zhang, J. (2018). Ridesharing query optimization using deep reinforcement learning. Proceedings of the ACM SIGMOD International Conference on Management of Data, 2141-2155. 41. Akhter, S., Awan, F. M., & Saleem, M. (2019). Machine learning techniques for query optimization in big data analytics. Journal of Big Data, 6(1), 1-25. 42. Chabot, Y., Goczyla, K., & Roussel, N. (2018). Using machine learning to optimize query plans in distributed databases. Proceedings of the IEEE International Conference on Data Science and Advanced Analytics (DSAA), 1-10. 43. Tahboub, R., Kalashnikov, D. V., & Mehrotra, S. (2017). Machine learning-based adaptive query optimization. ACM Transactions on Knowledge Discovery from Data (TKDD), 11(2), 1-20. 44. Sreedhar, R., & Kumar, M. R. (2020). Survey on query optimization using machine learning techniques. International Journal of Computer Applications, 176(1), 9-14. 45. Chu, X., & Ilyas, I. F. (2019). Machine learning for data cleaning and integration. Proceedings of the ACM SIGMOD International Conference on Management of Data, 2201-2205. 46. Wu, E., Krishnan, S., Kraska, T., Franklin, M., & Goldberg, K. (2016). Optimizing SQL joins using deep reinforcement learning. arXiv preprint arXiv:1908.03176. 47. Mozafari, B., Curino, C., Franklin, M. J., & Madden, S. (2013). Performance and resource modeling in highly-concurrent OLTP workloads. *Proceedings of the 2013 International 48. Garralda-Barrio, M., Eiras-Franco, C., & Bolón-Canedo, V. (2024). A novel framework for generic Spark workload characterization and similar pattern recognition using machine learning. Journal of Parallel and Distributed Computing, 189, 104881 49. Fathalla, A., Salah, A., & Ali, A. (2023). A novel price prediction service for ecommerce categorical data. Mathematics, 11(8), 1938 529 | P a g e