In case of a machine learning project, setting the right expectations is important. Most people have very limited or no understanding of machine learning and if you read magazine articles people seem to think machine learning is magical.

Well, that is not true. It is a proper engineering domain, so it is not simple. It is not going to work in five minutes, it usually does not work on the first try.

The most important questions are:

  • Why do you even do machine learning in the first place? 
  • What's the business problem you're trying to solve?
  • Why do you want to use machine learning?
  • Why do you think machine learning is a good technique?

Many people think it's the way to go because it's trendy. Not all problems can be solved with machine learning, so sometimes you have to use something else. But now let’s see a couple of uses cases, where Machine Learning was the answer, and it has a fixed a business problem.

Truckers in the United States cover more than 150 billion highway kilometers—enough to travel around the Earth over 3.7 million times. According to Convoy, a Seattle-based logistics company, these trucks are moving 10.5 billion tons of cargo annually. However, 40 percent of the kilometers truck drivers drive each year are done with an empty truck, causing waste of time and fuel. Convoy is trying to use machine learning to provide better matches for shippers and truckers, allowing them to move cargo more efficiently and reducing costs for both parties. Using Amazon SageMaker, they analyze millions of shipping jobs along with trucker availability, then recommend matches that are cost-efficient and timely. This influences routing, prices quoted to shippers and truckers, types of loads match best with individual drivers.

Kia Motors, South Korea’s oldest automobile manufacturer, builds more than 3 million vehicles a year. Kia is using computer vision inside the car to better understand and assist drivers. Using AI technology like Amazon Rekognition for advanced image and video analysis enables Kia to offer driver-assistance features like personalized mirror and seat positioning for different drivers behind the wheel. According to a report by the Center for Disease Control and Prevention (CDC), there are nearly 6,000 fatalities each year in the United States due to drivers falling asleep at the wheel. Driving under the influence (DUI) adds another 10,000 fatalities a year in the U.S.

Kia’s goal is to use AI technology to enable in-car DUI detection, as well as to monitor driver fatigue, to help decrease fatality rates. For instance, the in-car system may advise not starting the car at all if intoxication is detected or changing to autonomous-driving mode to safely pull the car over to the side of the road if driver fatigue is recognised.

GE Healthcare has adopted machine learning as a driver of better patient outcomes, with applications ranging from data mining platforms that draw on patient records to examine the quality of care to algorithms that predict potential post-discharge complications. They partnered with researchers at the University of California, San Francisco to develop a library of deep learning algorithms focused around enhancing traditional x-ray imaging technologies like ultrasounds and CT scans. By combining a variety of data sets —patient-reported data, sensor data and various other sources— into the scan process, the algorithms will be able to recognize the distinction between normal and abnormal results. According to a recent survey, 82 percent of healthcare decision-makers say the use this technology now resulting in enriched patient care, while 63 percent report lower readmission rates.

In 2018, the most popular TV program in the United States was football. NFL games accounted for 46 of 2018’s top 50 programs, averaging 15.8 million viewers over the season. Today, the NFL’s Next Gen Stats (NGS) program uses advanced tracking technology collected via RFID devices in the shoulder pads of every player and installed at each of its stadiums. These devices obtain data about which players are on the field at a given time, players' location within inches, and the direction and speed in which they move. Powered by Amazon's machine learning tool, SageMaker, the NGS platform allows the NFL to instantly and easily create and deploy machine learning models capable of understanding the gameplay. One example is Completion Probability metric, which combines more than 10 in-play measurements ranging from the length and speed of a specific pass to the distance between the receiver and the closest defenders—as well as the quarterback and nearest pass rushers. SageMaker helped decrease the time to get to results from as much as 12 hours to 30 minutes.

Capital One is one of the biggest banks in the United States and the largest digital bank. One area where Capital One is using machine learning is in the field of fraud detection. Some of the world’s worst cybercriminals concentrate on the financial services industry, making security all the more essential. According to a report from 2018, the White House council of economic advisors, malicious cyber activity cost the economy can be up to $109 billion in recent years, with the financial sector witnessing the most breaches of any industry.

Thorn was founded in 2012 by Demi Moore and Ashton Kutcher to battle online child abuse. This organization developed a product called Spotlight. Spotlight's advanced machine-learning capabilities save time for overworked, under-resourced investigators by automatically flagging advertisements likely to represent at-risk children and helping law enforcement agents not to get lost in a sea of online data.

Since 2016, Spotlight has supported officers in the United States and Canada open more than 21,000 trafficking cases and identify about 18,000 victims, including more than 6,000 children.

LinkSquares offers high-growth companies with a suite of tools to complete fast and orderly legal reviews of business agreements, consisting contract analysis and reporting, The tools they provide help relieve businesses from the burden of manually reviewing contracts. They use natural language processing (NLP), an artificial intelligence (AI) method assisting computers to interpret and understand human language, to build its machine learning algorithm. By running a machine learning solution on AWS, LinkSquares can concentrate its resources and time on developing new services and solving even greater business problems for their customers.

Saildrone gathers environment data with its wind-powered ocean drones. They use machine learning on AWS to better quantify the trends and behaviour of the main fish stocks and their predators, such as sharks. it helps with preservation efforts and sustainable fishery management. Saildrone just completed the first successful autonomous circumnavigation of Antarctica in 196 days, generating key insights into the ocean and climate processes, and the main area where used machine learning on AWS  was to identify the risk of collision with icebergs.

The average waiting time in the US for a parent who's worried that their baby or child might have autism is about a year. If a child is diagnosed at 18 to 24 months and provided two years of intervention, the average IQ gain is 17 points. Duke University has developed an app which can automatically detect autism and spectrum disorder. It uses computer vision analysis to assess facial expression and attention while watching a set of movies. While the child is watching these movies on either a smartphone or a smart tablet, the app is using the camera in the device to measure the child's response and it is able to estimate the chance of a disorder.

These are just a couple of examples where Machine Learning was useful. Whenever you are thinking about a Machine Learning project, even if you have a clearly defined goal and machine learning seems like a good answer, you will have to remember one thing: zero prediction error is never going to happen. So, you will always have prediction errors, your model will make mistakes. Zero errors in most Machine Learning uses cases are very unlikely, that's one of those expectations you need to set.

And why you should do it in the cloud?

The cloud provider’s pay-per-use model is good for bursty machine learning workloads and you can also use the power of GPUs for training without investment. It provides scalability. You can scale your model up to any degree to improve the performance. It also provides access to a cheap, unlimited storage.

Perhaps even more importantly, the cloud makes intelligent capabilities accessible without requiring advanced skills in artificial intelligence or data science—skills that are rare and in short supply. A survey by Tech Pro Research found that just 28% of companies have some experience with AI or machine learning, and 42% said their enterprise IT personnel don’t have the skills required to implement and support AI and machine learning.