When Do You Need a Machine Learning Model?

[MeganBowers]

Machine Learning (ML) just sounds cool—doesn't it? Maybe it’s because I’m a nerd, or maybe it is cool. Regardless, you’re probably used to hearing buzz around this topic. (Or at least seeing TV and movies that play with the concepts.) 

Source: GIPHY

If you are not from a data science background, you might want to avoid such a technical topic. But I believe it’s essential for people who work with data to understand when machine learning is needed and when it’s not.  

Whether you are a data analyst, financial analyst, BI developer, or consultant, you can understand the basics of machine learning: the use cases, the high-level processes, and the tools. And then, when you have data you want to get insights from, you will know which route to go down for that project. 

What is machine learning? 

Machine learning is a branch of artificial intelligence (AI) that focuses on using data and algorithms to create models that learn—i.e., improve accuracy over time as humans do. It sounds very complex (and it can be), but the simplest model can be a line of best fit through a plot of your data. This line of best fit has an equation that fits your current data and allows you to make predictions about future data.  

So when might a machine learning model be useful to you? This depends on what your goal is when working with a dataset. 

Define the goal of your analysis 

What do you want to learn from your data? Your answer to this question could fall into several categories: 

• You want to use your data to describe events or trends (descriptive analysis). You would like to know what happened. 
• You want to use your data to diagnose a problem (diagnostic analysis). You would like to know why something happened. 
• You want to use your data to predict future events or trends (predictive analysis). You would like to know what will happen in the future. 
• You want to use your data to determine your next actions (prescriptive analysis). You would like to know what to do in the future. 

If the end goal of your project is to predict or prescribe, then you need machine learning in some capacity. If your goal is to describe or diagnose, then machine learning may not be your first choice. You should likely start with some data analysis, and then you could turn to machine learning models for additional insight into things like variable associations. 

Machine learning use cases 

Think you might have a machine learning problem on your hands, but you’re not sure? Below are some examples of projects that have used Alteryx to create models to inform business decisions. 

Finance  

• Time series modeling/forecasting financial data (sales, pricing, stocks, cryptocurrency, etc.) 
• Predicting late payments 
• Forecasting insurance payments 
• Predicting the risk of default in AR 

Sales & Service 

• Predicting customer churn 
• Forecasting time to resolve complaints 
• Survey weighting 
• Customer segmentation 
• Predicting customers at risk for not renewing 

HR 

• Predicting worker attrition rates 
• Labor and staffing optimization 
• Modeling to uncover factors and sentiments affecting employee satisfaction 
• Optimizing employee learning and development 

Supply Chain & Manufacturing 

• Demand forecasting 
• Supply chain forecasting to optimize supply and inventory spend 
• Predictive maintenance 
• Optimize equipment utilization 

Marketing 

• Predicting customer defectors after marketing campaigns 
• Marketing campaign success modeling 
• Content recommendation engines 

Legal 

• Predicting litigation outcomes 

Misc. Industries 

• Healthcare: Modeling disease spread and predicting hospital shortages 
• Education: Predicting student retention 
• Travel: Prescriptive modeling for shortening connection times and developing multi-model air traffic courses 

Understand your data 

If you want to create a machine learning model, you need to understand your data first. As you work through cleaning your data and preparing it, you might uncover issues that would make the data unsuitable for modeling. You also might discover relationships between variables, imbalanced classes, null values, or a host of other things. All of these are good discoveries! This is the first step toward creating a model, and it is called Exploratory Data Analysis.    

The more you understand your data, the easier it becomes to move forward into modeling. If you are prepping your data in Alteryx, then try using the Data Investigation tools to get more insight into the data you are working with. If you are using Alteryx Machine Learning, then most of the Exploratory Data Analysis will be done for you (more on this tool later). 

Start with the simplest approach 

Once you know the goal of your analysis, understand your data, and have determined that you need machine learning to accomplish that goal, it is best to start with the simplest approach.  

Before you jump to creating state-of-the-art AI to solve your business problems, start with a simple, understandable model—linear regression, for example. If the accuracy of that model isn’t satisfactory, or your data does not work with the assumptions of that model, then move on to more complex regression techniques. 

Notice I used the descriptor “understandable” in addition to simple. Explainable AI is a hot topic because of the ethical and practical consequences of not understanding how your machine is learning and making decisions. And if you are just starting to venture into machine learning, it can get messy quickly if you have no concept of how your model works. 

So if you determine you need to create a model to solve your problem, and you are new to machine learning, where should you start? 

Evaluate your tool options 

Source: GIPHY

If you are brand new to machine learning, then the Alteryx Machine Learning cloud platform is a great place to start. It has a clean user interface that makes ML feel more approachable. As you progress through each step of your modeling, accompanying charts are automatically generated. And my favorite part is that you can turn on Education Mode, which offers in-depth definitions and explanations at every step so you can learn and understand the modeling process. 

The Alteryx Intelligence Suite is another great option if you have little to no experience with modeling. The Assisted Modeling tool will step through the process with menu options to create your machine learning model. While similar in some ways to the Alteryx Machine Learning platform, it is a tool in Alteryx Designer (rather than a platform), so you can do data cleaning and machine learning all in the same Alteryx workflow. 

If you know a little bit more about machine learning models, then the R Tools will help you create a model without coding. There are tool categories such as Predictive, Time Series, Prescriptive, and more. Each category contains tools for all types of models, as well as tools for comparing and evaluating your models. 

If you have machine learning experience, then you can jump into creating your model from scratch by using the R or Python tools in Designer. Or, you can use the “expert mode” in the Assisted Modeling tool. (Look mom—no training wheels!) 

Ready to learn? 

Are you ready to learn—or rather, teach those machines to learn? You can request an Alteryx Machine Learning Demo or sign up for an Alteryx Intelligence Suite Free Trial. We also have Interactive Lessons on Alteryx Machine Learning to help you jump-start your next project! 

If you have used Alteryx already for an interesting ML problem, comment below to share with the community.  

Additional Resources:  

Predictive Modeling Interactive Lesson 

Machine Learning for the Data Analyst 

Democratizing Machine Learning at McGraw Hill Education 

Why Alteryx Machine Learning? 

Special thanks to Minet Polsinelli (@minetpolsinelli), Chris deMontmollin (@cjdemont), and Neil Ryan (@NeilR)  for their help with this article.