Predictive Modelling or Predictive Analytics

Written by innote September 21, 2018 0 comment

Using machine learning for prediction is mostly referred to as predictive modelling or Predictive Analytics (PA).

The main reason why people usually associate with machine learning is because of its ability of predicting the future behavior of individuals, although, there are other problems and situations to which machine learning can be applied. All that is require is some unknown thing or event that you want to determine (predict), and this could be in the past, present or future.

Doctors examine their patients, carry out tests and question them about their symptoms in order to gather (data).

Machine learning can be applied in the same way. Machine learning can be used to estimate the probability that someone has a certain condition, based on the symptoms that they present through the means of giving a host of detailed information about the symptoms of different illnesses. Another way to think about machine learning/predictive analytics is as a method of reducing uncertainty.

Despite the fact that there are a whole host of possible outcomes that could occur in any given situation. Machine learning won’t tell you with absolute certainty which outcome will occur, but it can provide some insight into the likelihood, or odds, of each outcome.

The output that is generated by the machine learning process is called a predictive model. The model help to captures the relationships (patterns) that have been uncovered by the mathematical process. It can be used to generate new predictions once a model has been created. Regardless of the type of model or the mathematics used to create it, a model’s predictions are almost always represented by a number – a score.

A machine learning is a process while the predictive model is the end product of that process.

Machine learning can be applied in all sorts of situations and too many types of problem. The higher the score the more likely someone is to behave in the way the model predicts, the lower the score the less likely they are to behave in that way.

A good AI application is one that can perform as well or better than the average person when faced with everyday tasks. A key mistake to avoid is thinking that current AI applications are in any way intelligent in a human conscious way.

The core components that drive most AI/machine learning applications are:

Data input. This can be sensory inputs from data series, cameras, microphones or other sources.

Data (pre)processing. The raw data input must to be processed into a standard “computer friendly” format before it is ready to be used. Also a big task is to wash the data for errors.

Predictive models. These are generated by the machine learning process using past experiences; i.e. large amounts of historic data. Pre-processed data for new cases is fed into the models in order to generate fresh predictions going forward.

Decision rules (rule sets). A prediction on its own is useless. You have to decide how to use it. Decision rules are used in conjunction with data inputs and the scores from predictive models to decide what to do. Sometimes these rules are derived automatically by the machine learning algorithm, but often they will include additional rules defined by human experts/business users.

Response/output. Action needs to be taken based upon the decision(s) that have been made. If the decision is that someone should be hired, then they need to be sent an offer letter, given a contract to sign and so on.

AI is the result of the combination of these individual components. Sheer complexity one of the algorithms that underpin them is responsible for some AI applications to appear cleverly, combined with a slick user interface to gather data and deliver the required responses in a human friendly way. integrate these components into cars and other vehicles, or combine them with the latest generation of industrial machinery,

Combine these components with the latest generation of industrial machinery, and one has robots that can interact with their environment and engage with us in a very human like way.

As mentioned, the vast majority of AI applications/machine learning which is being used in business rely mainly on predictive models. It is these models that generate numbers (scores) which help to indicate what is likely to occur, according to the information that the model is presented with.

The predictive model scores can be generated in two types which are:

Probability (likelihood) scores: This is the process in which the likelihood of some specific event are been predicted. The technical name for a predictive model of this type is a classification model.

Magnitude (quantity) scores: These is the process of predicting the size or the amount of something. The technical name for a predictive model of this type is a regression model.