Machine Learning can often be overlooked or ignored by businesses for a number of reasons – perhaps because of the perceived time and effort that would be required to implement it or because they don’t know how to make the best use of it. When fully embraced however, machine learning can be a highly effective tool to assist decision making that can help get an edge over the competition and identify areas of improvement. 

Machine Learning models

Any attempt to develop a machine learning model for a problem should begin by attempting to use very simple models, as these are most often the easiest to explain. For example, a linear model could be used, which has a high degree of explainability due to the fact that the co-efficient of each feature essentially gives the importance of this feature in the decision-making process. These models are explainable.  

If the performance of linear models is not great enough, then a non-linear model or Generalised Additive Model could be used. These have better performance but have less explainability as you can no longer simply look at weights or co-efficients as a measure of importance.  

If even greater performance is required, there are more complex models such as XGBoost or Deep Learning. As these models are inherently difficult to determine reasoning for decisions from, you would have to determine some form of importance of features in decision making using alternative methods:  

• Shapley Values – Treats the Machine Learning algorithm as a ‘black box’ and determines from varying the inputs one feature at a time, which features contribute more to the output, essentially giving us a value representing the importance of a given feature. For example, the feature “Number of Previous Insurance Claims” might contribute most to the decision not to offer insurance. 
• AWS Clarify – Can give insight into decisions made by machine learning and increase explainability, along with detecting potential bias during data preparation and after model training. More can be read here.

Machine learning uses

While some models are easier to explain than others, there are a number of ways that machine learning can be implemented to improve business processes. 

• Predictive modelling is a process in which data and statistics are used to predict outcomes with data models. It can be used to study data and predict what is going to happen next and in turn inform what decisions should be made next. 
• Process mining evaluates business processes and can be used to understand where processes have gone wrong and find new ways to make your business more efficient. For example, process mining could analyse data from previous sales to tell you whether or not you are likely to run out of stock.  

Having Machine Learning in your decision making can allow your business to identify new areas of improvement and make vast enhancements to the way you conduct your business. What’s more, machine learning can work away in the background so you don’t have to dedicate precious time to maintaining and monitoring it.  

There are a number of different ways to use machine learning to help your business decision making and Pulsion are experts at delivering custom machine learning solutions. Get in touch today for a free 30-minute consultation where we can chat about the solutions that are right for you.