5 Ways Businesses With No Data Can Use Machine Learning

Source: Deep Learning on Medium


Artwork by Mariah Arvanitakis

At Northraine, often we get approached by businesses that really want to use machine learning, but they don’t know how. They either have very little data on hand or their data is all over the place and lacks predictive power. Luckily, like most roadblocks, there are ways to get around this. Below, we’ve listed 5 ways to use machine learning for businesses that may not have the sleekest dataset in town (or any data at all) but long for the power and insight that AI can provide them.

  1. Depending on what you’d like to achieve with machine learning (this is an important question to ask), you may not even need that much internal data at all. Sometimes the most creative projects start from information that was available all along, just waiting to have a model built from it! There is a wealth of data sets out there, many are free and publicly available. From sommelier ratings of Italian wines to recordings of abnormal heartbeats used to recognise heart murmurs — the Internet is awash with exciting information just waiting for a machine to learn from it.
  2. As well as these niche datasets, the government and its various departments often make available incredible sources of knowledge on everything from traffic to the ageing population, and at times even encourage us to do something with it. Australia releases all kinds of reports that if analysed properly, can bring real value to a lot of businesses and NFPs. Some interesting reports released recently cover topics including healthcare, flight patterns and road safety, just to name a few. Machine learning can turn these static reports into something really powerful.
  3. Social media is an incredible source of memes and wisdom. If there’s one way to detect consumer sentiment, public opinion or to get a hold of what’s important to ‘the people’ it’s scraping social networks. Instagram is incredibly powerful for predicting trends and emerging products. Twitter can be a bit more serious, so much so that one of Northraine’s brilliant data scientist’s — Yeshna — has just completed her thesis looking at how Twitter users use words like ‘asthma’ and ‘hay fever’ and conducted semantic analysis to build a predictive model which can provide early warnings for a public outbreak of asthma.
  4. ChatBots can be programmed through Natural Language Processing (NLP) with relatively less data than other machine learning applications. If a business has a good grasp on their customer’s pain points and FAQs, we can build a ChatBot or voice assistant that takes the pressure off their customer service team so they can focus on bigger and better things. ChatBots are automating everything from paying bills and selling your car to ordering a pizza.
  5. Last, but certainly not least, deep learning. Deep learning technically does need data, but it doesn’t need to be organised in a succinct way like a traditional machine learning training dataset. In fact, deep learning can be applied to just about any kind of unorganised data, collections of images or a myriad of different variables to find correlations and insights that a business might not be aware of yet. Deep learning can make blurry images clear, generate an impression of what far galaxies look like and translate content into multiple languages!

So next time you’re dreaming of using machine learning, but maybe feel you don’t have the right data on hand, just remember there are ways around this hurdle.