Machine Learning Primer

Machine learning (ML) tools are being broadly used in a variety of applications and tools to help businesses upgrade and drive their business decision-making to a higher level. Machine learning technologies are already in the market and are proving themselves as key tools for better decision making regardless of the size of the business or even the amount of data. ML is fast becoming a requirement for businesses to compete in the modern marketplace.

What is Machine Learning?

Machine learning is the ability of computers to learn rules from recognizable patterns in data perform tasks on their own. The data can come from a variety of sources, for example customer support tickets, health data, crime records, etc. Based on it’s learning from the patterns in the data, ML system arrives at a conclusion about expected outcomes such as customer issues, crime activity, or health outcomes. Humans help the machine’s learning process by providing sample outputs from which the ML derives algorithms, or sets of rules, which it can then apply to new data to get additional or new outcomes. Therefore, ML does not need to be given a precise set of rules and ML does not solely following the algorithms the programmers create. ML is trained and it finds the rules that would determine the various outcomes. ML then feeds the patterns and algorithms that it has identified back into the applications to identify insights from data. In this way, ML can continuously learn from a stream of incoming data and continue to generate insights with greater accuracy as the volume of the data grows. The field of artificial intelligence, of which machine learning is a part of, is the science of automating human thinking.

Machine Learning Applications

Machine learning is not a new concept, but today it is feasible because of the amount of data available and ability to resize computing capacity in the cloud.

Depending on the users’ needs, ML can perform a number of tasks with the data that may not be immediately visible to the human eye or may take too long for humans to identify. ML can be used to be able to classify the data, creates forecasts based on historical data, group data, or identify outliers. In

Data Classification: Data classification is used for binary models such as yes/no, approve/fail, etc, such as loan approval or for multiple classifications of the data into one of multiple categories such as identifying multiple plants in an image.

Regression: Regression takes into account multiple factors to forecast an outcome such as property prices or treatment results.

Data Clustering: Data is clustered into different groupings based on features in the data.

Anomaly identification: looks for datapoints that don’t fit the unusual pattern such as credit card fraud.

Challenges of ML

For ML to learn, it needs good samples that correspond to the desired outcome. For example if we want ML to identify parrots, we would need photos that would be good examples of parrots so the ML can learn what to look for in identifying a parrot. So the quality of the outcomes from ML depends on the quality of inputs into the system and represent the expected outcomes well. Similarly, ML requires a quantity of quality data to learn from with accuracy. Increasing the volume of good data available to ML improves its training. Smaller size data can be used but may require additional training to improve the outcomes.

Once the ML system generates results, those results must be reviewed to determine how well the outputs match expectations. Feedback from reviewing the results reinforces the learning so that it can generate better results. The causes of suboptimal results may be data issues, poor representation of the data, or lack of SME knowledge to provide quality feedback for learning.

The ML systems requires people to set it up and run it. ML applications can have built-in biases that could affect decision outcomes

Benefits of ML

Improvements in business decisions making that result from employing ML with pay off in a short time. Machine learning can make immediate and tangible improvements business decision making, improve products, and gain greater customer satisfaction.

Existing AI tools, such as Azure Machine Learning make it easier to build ML tools and get results relatively quickly. The tools are complex, but can be applied to solve problems for businesses of all sizes.

Time savings: ML-based applications can automate time consuming activities and free up SME time to do higher order activities. Turnaround time for decision making can also be shortened as the ML application can undertake routine activities.

Cost Saving: Costs savings result since the applications can help make lower level decisions freeing up people’s time to focus on more value added tasks.

Machine Learning Use Cases

Following are examples of use cases for Machine Learning.

  • Credit risk
  • Crime prevention
  • Customer support issues
  • Disease Detection and Progression
  • Facial Recognition
  • Fraud detection systems
  • Language translation
  • Make selection recommendation
  • Predict customer churn
  • Predict student dop out rate
  • Predictive maintenance
  • Sales forecasting
  • Sentiment analysis
  • Shipping, logistics and transportation
  • Understanding spoken language
%d bloggers like this: