Machine learning uses a powerful set of iterative algorithms to systematically examine large data sets, extract the fundamental patterns and interrelationships, and make predictions to new data.

What is Machine Learning?

Statistical modelling seeks to understand the structure of data but requires the data to align with certain assumptions and theories to be most effective.

Machine learning (ML) takes advantage of advances in computational power to probe for patterns in data without assuming the data meets certain conditions. ML algorithms are efficient at analysing large volumes of historical data, identifying the interrelationships between variables and using this information to predict future outcomes.

What is driving ML growth?

The ML explosion across many sectors is principally driven by an increasing abundance of data in combination with advances in computational power and low cost, scalable cloud computing. The theory underpinning most ML algorithms was developed decades ago but improvements continue to be made.

ML is used in widely disparate situations including business analytics, online search, cyber security, fraud detection, face and voice recognition, autonomous vehicles, robotics, virtual assistants and disease detection.

How can ML help your business?

  • Identify cases requiring early intervention
  • Red flag high cost situations
  • Fast track low-cost cases
  • Identify potential cases of fraud
  • Pinpoint areas of cost leakage
  • Uncover best practice approaches
  • Diagnose problem areas

Which ML algorithm?

Since 2013, we have used a proprietary suite of ML algorithms provided by a US based, award winning leader in machine learning. Their predictive analytics suite includes algorithms for decision trees, gradient boosting, automated predictive regression, and random forests. Their board of scientific advisors includes leading academics in statistics, machine learning, data mining and biostatistics.

The algorithms in this suite are:

  • Highly predictive – their algorithms have been used to win many high profile data mining competitions
  • Comprehensive – models allow for interactions between variables, capture non-linear variable relationships
  • Resistant to over-fitting, outliers and missing values
  • Have built-in performance improvements to the ML algorithm which are not available in open source modelling software like R and Python

What is a typical ML Workflow?