Data science is becoming crucial for making quick data-driven decisions these days. And a significant component of modern data science is machine learning (ML). ML models can identify trends in data and imbue an app with powerful artificial intelligence (AI) capabilities. While the benefits of ML in application development are readily apparent, machine learning experts are hard to find, and there is a high demand for data scientists who can create AI/ML solutions.
Some folks are turning to low-code/no-code solutions to rectify this AI/ML talent gap. Tooling in this area ranges from open-source libraries to end-to-end platforms that enable you to collect data, develop models, and test them. Many of these low-code platforms provide drag-and-drop capabilities, which could help automate tedious processes for experienced data engineers and democratize AI for less technical users.
If you can program an idea from scratch, there’s likely a low-code/no-code solution or API willing to do the heavy lifting for you. And, the world of data science is no different. Below, we’ll consider how low-code/no-code can be helpful in data science, digging into some benefits and potential use cases. We’ll also cite some example platforms and libraries so you can get your hands dirty with low-code AI/ML on your own.
The Need for Low-code Data Science
Data analysis is fundamental for a successful enterprise in 2022. 81% of leaders agree that data should be at the heart of all decision-making, according to a global report by EY. While most organizations desire to be data-driven, getting there isn’t all that easy — much effort is required to construct actionable data workflows.
This is partially because applications are generating high amounts of unstructured data. In fact, multiple industry analysts estimate that 80-90% of all the data in an enterprise is unstructured. These siloed data lakes consist of video, text, audio, social media data, and so on. Applications produce logs in various styles, too, making it cumbersome to standardize different formats before they can be utilized in ML applications.
Another aspect is the sheer complexity of developing ML/AI applications. Ideally, as a business generates greater amounts of data, it could analyze trends to better understand the impact of features and refine customer experiences. Most digital apps can also benefit from AI abilities like sentiment analysis or image classification. Creating useful ML models, however, requires skill. It requires data collection, data cleansing, model training, feature engineering, exploratory data analysis, and other advanced processes. Indeed ranks machine learning as the top in-demand AI skill, meaning access to such talent is hard to come by.
Uses of Low-Code in Data Science
With all that in mind, here are some ways low-code/no-code can help data science:
Ease the data collection process: First off, low-code could help integrate with APIs to seamlessly aggregate data from various sources. This could be an internal database or a third-party SaaS system. By utilizing pre-built connectors, you can more easily increase the number of data sources, too, thus improving your algorithm. Programmatic, automated connections become more critical when acting upon data in real-time.
Help cleanse and prepare data: As noted above, most of the world’s data is unstructured and will require clean-up to grant actionable insights. No-code automation could be leveraged to cleanse data and prepare various sources in a format suitable for ML training. Data type transformations may also include matching, sampling, shuffling, and scaling. “Finding effective ways to bring that data together under a more unified schema or format is critical,” said Maxim Wheatley, a member of the executive leadership of Merico.
Provide no-code AutoML to amateurs: Many libraries and no-code platforms can train an algorithm based on raw data. Utilizing such platforms could democratize powerful machine learning capabilities. No-code AutoML platforms include Google’s Cloud AutoML, Ludwig by Uber AI, Baidu’s EZDL, and Obviously.ai. “Open-sourcing and no-code AI platforms serve these companies’ broader goals of staying at the cutting edge of technology,” explains AI consultant Alexandre Gonfalonieri.
Accelerate the training, testing, and deploying of ML: TensorFlow is a well-known open-source utility for deep learning, but it certainly takes time and code to implement. Plenty of other low-code libraries can assist data scientists with different aspects of training and deploying ML with less code. PyCaret, for example, provides end-to-end ML model creation. Other tools include Auto-ViML, Apple’s CreateML, RunwayML, and Teachable Machine. Some may require knowledge of Python or R to fully leverage, while others are more codeless.
De-silo data and business insights: Low-code/no-code platforms enable data science with many pre-built components in an easy-to-use interface. In effect, the tools could help bridge the gap between data science and business units. This increased collaboration between departments could help improve end business outcomes. “Smart teams are figuring out more quantitative ways to define and assess success so that they can scale and double-down on what’s working,” Wheatley.
Lower the bar to data organization: Drag and drop interfaces could greatly help organize and structure data with usable flows. “[Low-code] might introduce ‘plain-English’ scenario-driven or question-driven queries to focus on the outcomes and goals of a data query, rather than the process to get there,” explains Wheatley.
Generate dashboards and reporting: Beyond ML/AI preparation, low-code can help data scientists generate visualizations of their data. This could aid quarterly reviews or help audit an organization’s data footprint. Many low-code development platforms provide modules to generate sleek user interfaces and graphs based on a dataset.
Final Thoughts: Low-code And Data Science
Low-code AI/ML can empower knowledge workers in countless roles to “solve their own problems.” For example, using a low-code tool provided by KNIME, a marketing analyst was able to assemble a sentiment analysis natural language processing (NLP) solution in a matter of days. Image classification is essential for content moderation initiatives, and it can empower quality engineers to automatically discover product defects on the manufacturing floor. AI/ML applied to anomaly detection is also becoming more important in the financial sector to detect anomalies and prevent fraud.
Gartner predicts that low-code development will comprise 65% of new applications by 2024. While low-code can bring significant agility benefits, it’s essential to consider the limitations of such tools and the downsides of too much automation. “It is safe to assume that your data scientists will not feel comfortable using no-code/drag and drop tools,” writes Gonfalonieri. Many data scientists are comfortable coding, and may feel constricted by a no-code development platform.
Though low-code lowers the bar of entry, “in many cases, low-code isn’t going to replace or eliminate the need for specialists or technical contributors,” added Wheatley.
With that being said, AI is becoming mainstream, and low-code tools can help slim down requirements to automate many aspects of complex ML projects. “Low-code is becoming increasingly adopted by many industries, as it provides vast opportunities for innovation and experimentation while maintaining lower costs,” said Jonathan Grandperrin, CEO of Mindee. Adds Grandperrin:
“Professional developers will get time back from some of the more mundane programming activities, such as data extraction, therefore speeding up the creation of apps for commodity functions and spending more time on enterprise-class apps that still require higher programming skills.”
