The focus of our Computer Vision blog series to date has been focused on understanding the limitations, benefits, and complexity of applied computer vision. Machine learning is a broader field that encompasses numerous methods, including computer vision and more organisations are beginning to see the value in its application to their day-to-day operations.
However, I often come across the refrain of “we want to apply machine learning, but we don’t know where to start.” If you have had this thought, you’re not alone. The possibilities of these methods appear endless and starting on that journey can seem impossible; especially when you consider something as complex as computer vision. While plenty can be said about the role data management plays in the successful adoption of machine learning, it can be useful to focus on cases about how data science can be applied.
- Exploratory Data Analysis. While this is not strictly a machine learning task, it is an essential step in the process that can have a lot of value on its own. When embarking on a machine learning project, the first step in this process is to explore and understand the data. The outcome of this analysis is an understanding of data quality and format, potential outliers, and correlation between variables. Have you ever been supplied with a spreadsheet that has thousands of rows and you just want to understand the data? Are you worried that there might be data missing? How do you identify values that are unusual or may be incorrect?
- Text Recognition. It is not uncommon for people to spend large chunks of time digitising text from handwritten forms, printed text, PDFs, or image files. These time-consuming data entry tasks can make use of techniques like optical character recognition (OCR) to automate the process. While this is common practice for many people, others remain unaware of it and the time it may save them. How do you process receipts in your business? Through automation or by hand? Perhaps your team writes inspection reports by hand and those notes are then scanned and stored digitally, while effective, this process does not allow you to electronically search the document. By applying text recognition, you can turn those documents into searchable records.
- Basket Analysis. Also known as affinity analysis, basket analysis takes historic sales records and analyses them to understand the relationship between different products or services. We can examine transaction information to determine which products are sold together. For example, is a customer that buys milk more likely to purchase eggs? The findings from this type of analysis are invaluable when it comes to cross-selling and understanding patterns in customer behaviour. What would be in the basket for your company? We all get a feel for our products and services, but do you have a way to quantify your basket beyond experience?
- Customer Segmentation. It is one thing to understand how your customers interact with your service, but it is another to understand how your business should interact with customers. Customer segmentation splits your customers into groups based on similarities. Many of us will have seen this in action when it comes to targeted advertising. These techniques are not limited to individuals but can be applied to organisations too. Businesses may be grouped based on numerous factors such as industry, invoice records, or location. The groups can then be leveraged to strengthen customer relationships by developing more tailored communication and marketing strategies as well as understanding who your potential customers may be. Do you offer loyalty discounts to your engaged customers, who are unlikely to leave, or to the next group of customers that you want to become engaged? Do you know which customers are less likely to return?
At its core, machine learning is a tool that can be used to reduce time and cost while improving quality and outcomes. If you spend time on pricing strategies for various products and services, perhaps dynamic pricing may be of the most interest. However, machine learning solutions don’t have to be end-to-end. Using dynamic pricing as an example, a machine learning model could be used to recommend an appropriate pricing range or assess your price against the market and leave the final decision to a human being. It asks the question, what’s something you do that’s time consuming or what insights would be beneficial to your day-to-day work?
If you want to discuss these use-cases further or have a problem that you think could be solved with machine learning, touch base with Joe Duncan.
Computer vision blog series:
Blog 1 - Data Science Researcher joins Abley
Blog 2 - Using computer vision to detect traffic signs
Blog 3 - Data labelling
Blog 4 - The realities of "big data" in computer vision
Blog 5 - Selecting the right data for computer vision
Blog 6 - Non-existent cats and other data augmentation magic
Blog written by Joe Duncan, Data Science Researcher