Machine Learning for Reengineering Business Processes

The review thus far has focused on available frameworks, processes, tools, and approaches for manual business process re-engineering, as well as success and failure criteria. As a result, the next stage is to look at the use of machine learning and data mining. The evaluation looked at 60+ papers on topics like process management, process improvement, process re-engineering, process optimisation, process automation, process visualisation, process modelling, process planning, process discovery, and process behaviour.

 

According to the research, academic interest in process science, process mining, data mining, and machine learning is on the rise. This indicates a significant emphasis on process behaviour, process improvement, and business process management throughout time. The study does, however, show a movement in focus from understanding process behaviour and business process management in 2017 to process improvement and optimization in 2020, which is a normal progression of research focus from broad to narrow research concerns. Furthermore, machine learning’s developing skills have a promising future and can be applied to more complex problems.

 

Tools and Techniques

Furthermore, both frameworks and methodologies use tools and procedures throughout the process, according to the review. Also, the review discovered a variety of tools and methodologies that have been employed in business process re-engineering for various goals. There are tools for process discovery and process visualisation that allow you to get a bird’s eye view of the flow of the process from one activity to the next, as well as an overview of the roles and responsibilities. Monitoring the quality of the business process or analysing characteristics of the process such as process performance, reliability, and efficiency are some more techniques for business process management.

 

Therefore, in addition to their impact on process improvement and reengineering, systemised tools and approaches have a practical application in generating, analysing, and visualising data throughout the process.

Furthermore, the review concludes that the use of such technologies is critical for ensuring the success of implementation initiatives.

 

Success Factors

Any implementation team’s ultimate goal is to have a high success rate in implementation projects. Business process re-engineering projects, on the other hand, were not always successful for a variety of reasons, the most common of which was the use of best practises or industry standards in an industry field based on other industrial experiences without adequately studying the field’s unique requirements. Furthermore, due to a lack of a solid framework or methodology adaptation, business process re-engineering implementations fail at a rate of roughly 70%. However, a number of elements influence the success or failure of a project. They can be used as indicators to anticipate the outcome of a project’s implementation and the percentage chance of success.

The researchers added a category level to classify the factors into driver, strategic, or enabler categories. The driver factors derive the need for change and raise the flag in the organisation when the business process requires improvement. The strategic drivers direct and steer the project to implementation. The enabler factors are necessary to enable the successful implementation. 

 

Machine Learning as a Predictive Process Monitoring Technique

ML is an artificial intelligence branch that employs real-world knowledge to make human-like judgments without the need of pre-defined rules . In a semi-automated approach, ML employs statistical methods to learn structural patterns in often enormous datasets . Classification, grouping, regression, and anomaly detection are some of the most common applications of machine learning.Furthermore, supervised and unsupervised learning are two types of machine learning methodologies. Supervised learning employs previous data that has already been classified by an outside source to replicate classifiers. Unsupervised learning algorithms, on the other hand, process data on their own to gain insights. supervised learning is appropriate when categorising cases according to established classes, but unsupervised learning is typically used to group comparable cases with an imprecise definition of classes e.g., for anomaly detection.With predictive process monitoring, possible outcomes of tasks and instances or predefined business goals can be used to specify relevant classes. Thus, predictive process monitoring belongs to the field of supervised learning.

In supervised learning, RF, logistic regressions, SVM , and shallow neural networks  are all traditional ML techniques . The fact that the underlying algorithms are simple for humans to understand explains why these strategies are so popular. In recent studies on outcome-oriented predictive process monitoring, the use of RF in particular has become commonplace . RF is an ensemble learning technique that consists of a forest of decision trees. Each tree in a forest forecasts a class, and the final classification is based on the class that receives the most votes . When using single decision trees, RF avoids overfitting by doing so. By maximising the smallest distance, SVM aims to discover a hyperplane that divides data into two classes. Making use of the “kernel trick” (i.e., solving the problem in higher dimensions)

The format of input data has a big impact on how well standard ML techniques work . If the data is preprocessed in a way that aggregates basic features into features with increased information richness, RF gives significantly better outcomes for strongly correlated features in particular. Feature engineering is another term for this type of preprocessing . By identifying both the mapping of features to labels and the most essential feature combinations, i.e. a high-level structure, representation learning tries to automate human feature engineering  High-level representations are challenging to find in circumstances when input attributes are highly interconnected .

 

Machine Learning Classifiers’ Performance Evaluation

When developing ML classifiers, it’s critical to evaluate their performance. For each data point, the classifier’s predicted class must be compared to the actual class. A “true” forecast is one in which both classes (predicted and actual) match.

Several metrics can be calculated using this foundation. Accuracy provides a convenient assessment by collecting all forecasts into a single score. However, in the situation of unbalanced data, using accuracy as a single evaluation criterion can lead to deceptive conclusions. If there are 95 normal and 5 problematic examples, for example, a classifier can achieve a very high accuracy of 95% by categorising all instances as normal. However, we would not consider a classifier that never raises an alarm in the event of a problem to be performing well.

 

Business Process Re-engineering in ML

 

This research shows that business process re-engineering is a well-established scientific area. Because of the amount of data collected and analysed during the re-engineering process, the study discovered that business process re-engineering is equipped with numerous strong frameworks, methodologies, tools, and approaches. The data is extremely useful for analysing process performance, scientifically determining the need to re-engineer a business process, and determining the best re-engineered business process for changing business requirements.

 

There were also limited attempts to use machine learning for business process reengineering, according to the analysis. This demonstrates that artificial intelligence and machine learning can help academia contribute more to business process reengineering automation.

 

This research provides a novel way to automate business process re-engineering based on key Lean Six Sigma methodology concepts, which are developed from the basic concept of minimising waste and variations.

 

The researchers plan to test and qualify the proposed approach in additional case studies in the future. In addition, Brisk Logic intends to create a platform that will be open to the public for experts to exploit and challenge.