When to use Machine Learning
Posted on June 24, 2022 Big Data Machine Learning & AI
Machine Learning (ML) and Artificial Intelligence (AI) are at the forefront of many technologies. ML and AI models can perform exceptionally well at a variety of tasks: not only can they automate tasks, improve efficiency and performance, but they can also help you do things you were not capable of before. But how do you determine if an AI or ML model will be of benefit to your process?
Some situations where ML models can be highly effective include the following:
- For prediction and classification tasks. Do you want to predict performance outcomes or provide a feedback control? Do you want to classify anomalies or identify equipment in need of maintenance? Then ML models can be beneficial.
- If you cannot write down a simple set of rules that control your process. ML models are exceptional at finding complex patterns in data.
- The scale is large. Manually processing a few cases is manageable, but as the data rate and quantity of data increases, manual management becomes tedious or impractical. ML models can efficiently handle large scale data. They can help identify which data is the most significant and the best ways to summarize this data.
- When you have unstructured data or high dimensional data such as images, videos or multiple timeseries. For tasks involving these types of data, ML models almost always produce state-of-the-art performance.
One final thing to bear in mind: there is no one model that performs best on all problems. Different model architectures are each suited for diverse types of problems, so the choice of which ML model to use is important.
With that said, there are times when you may not want to use ML. Examples of this are when your data has certain limitations or gaps, when you want to have an easily interpretable model, or if you only want to look at historic trends. In these cases, more traditional data analytic methods might be more appropriate. We work closely with our client to determine if ML is a good option or whether other mathematical or statistical analysis options are more appropriate.