Wednesday, September 3, 2014
Machine Learning in Forecasting
In the 2011 movie, Moneyball (adaptation of Moneyball: The Art of winning an unfair game by Michael Lewis), Brad Pitt plays the role of a General manager who attempts to assemble a competitive team by using techniques of statistical analysis. The manager uses metrics such as on-base percentage and data analysis to make better decisions in selecting the right players. The Oakland Athletics won 20 consecutive baseball games, highest number of wins in a streak in American League history. In the recent 2014 FIFA world cup, companies such as Google, Baidu and Microsoft used the data analytics techniques to predict the winners at each stage and the success rate of their predictions was pretty impressive.
The underlying field of study which these analysts are relying on is Machine learning and is a growing field whose applications we can see in every walk of our life. Machine learning is “construction and study of systems that can learn from data”. Machine learning is being used across many domains and functions. For example, Facebook uses these algorithms to recognize faces in photos and Netflix’s uses these algorithms for its recommendation engine.
There are several types of machine learning results – the main ones are:
Regression – This is achieved using the real values. For example, if a seller wants to estimate value the cost for his house, using machine learning methods, we could estimate a price based on the old data.
Classification – This is classified for discrete values. – Discrete values rely on limited results like win or loss.
Supervised – Supervised learning finds use in most of the modern technologies such as recommendation engines or facial recognition software.
Unsupervised – This type of process has applications in fraud detection, genetics analysis and finance.
Let us take this example and apply the concepts to apparel industry to understand demand forecasting. In order to develop the right architecture based on extreme machine learning concepts to deal with apparel demand forecasting, the problem needs to be decomposed and analyzed.
With the increasing emphasis on Big Data, a lot of machine learning techniques in time series forecasting are focusing on three aspects: the formalization of one-step forecasting problems as supervised learning tasks, the discussion of local learning techniques as an effective tool for dealing with temporal data and the role of the forecasting strategy when we move from one-step to multiple-step forecasting.
According to a report issued by the IBM Institute for Business Value Chief supply chain officers plan to invest in analytics, business intelligence, and other software tools to bring more visibility to their supply chains . The study reveals that 92 percent of the supply chain executives said that they expect to use advanced analytics using machine learning in the next two to five years. Additionally, 62 percent said that in the next three to five years they plan to invest in software and tools that will help them achieve supply chain visibility.
Will Machine learning cause a revolution in the Supply chain industry like it is doing everywhere else?