Predictive analytics

Predictive analytics models are designed to assess historical data, discover patterns, observe trends, and use that information to predict future trends. Popular predictive analytics models include classification, clustering, forecast, outliers, and time series, which are described in more detail below.

Classification models
Classification models are categorized under supervised machine learning models. They place data into categories based on conclusions from the historical data. This model is commonly used for answering questions with binary outputs, such answering yes or no or true and false. Types of classification models include logistic regression, decision trees, random forest, neural networks, and Naïve Bayes.

Clustering models
Clustering models are categorized under unsupervised learning. They sort data into groups based on similar attributes. For example, an e-commerce site can use the model to separate customers into similar groups based on common features and develop marketing strategies for each group. Common clustering algorithms include k-means clustering, mean-shift clustering, density-based spatial clustering of applications with noise (DBSCAN), expectation-maximization (EM) clustering using Gaussian Mixture Models (GMM), and hierarchical clustering. 

Forecast models
Forecast models use metric value prediction, estimating numeric value for new data based on trends from historical data. For example, a call center can use the model to forecast how many calls it will receive per hour. Time series and econometric models would be examples of forecasting models.

Outliers models
Outliers models deal with anomalous data entries in a dataset. For example, insurance companies can use it for fraud detection to flag anomalous data within a list of transactions.  Some popular methods for outlier detection include extreme value analysis, probabilistic and statistical modeling, linear regression, proximity-based modeling, and information theory modeling.

Time-series models
Time-series models employ a sequence of data points using time as the input parameter. It can take the last year of data, calculate a numerical metric, and use that metric to predict the three to six weeks of data. For example, the model can be used by a hospital to make predictions about emergency room capacity based on the number of patients who showed up in the past six weeks.

Predictive analytics begins with a business goal, such as to reduce waste, save time or cut costs. The process uses models to harness massive data sets to generate outcomes that support that goal.

As an example, the predictive analytics process for predicting sales revenue follows these basic steps.

1. Import data from a variety of sources. These data sources include product sales, marketing budgets, and national GDP.
2. Clean the data by removing outliers (i.e. data spikes, missing data) and aggregating. A single table could be used to aggregate different types of data, such as product sales, marketing budgets, and national GDP.
3. Develop a predictive model based ensuring appropriate fit. For example, neural networks could be used to build and train a predictive model for the revenue forecasting.
4. Deploy the model into a production environment, where it can be accessed through other applications.

The use of predictive analytics is a key milestone on your analytics journey — a point of confluence where classical statistical analysis meets the new world of artificial intelligence (AI). Today’s unprecedented convergence of intuitive tools, new predictive techniques and hybrid cloud deployment models makes predictive analytics and modeling more accessible than ever before. For the first time, organizations of all sizes can have the tools to embed predictive analytics into their business processes and to harness AI at scale.

The evolution to an enterprise data science program can deliver significant competitive advantages. The typical steps in that evolution are:

Phase 1: Getting started

When a business begins building its data science capabilities, it usually starts with ad hoc projects, such as developing models to answer specific questions or support research projects. With solutions such as IBM Watson Studio Desktop, data scientists can work 24x7 on their own computers or laptops and sync up with a wider team when needed.

Phase 2: Adoption growth

As data science is adopted more widely across the business, different departments need to deploy their models, connect them to data sources and infuse them into production applications. IBM Watson Studio and IBM Watson Machine Learning make it easier for departmental data science and IT teams to collaborate across this lifecycle.

Phase 3: Enterprise-scale adoption

Once AI is embedded into business-critical processes, organizations need to build a central platform to manage and govern models and data. IBM Cloud Pak for Data can provide the infrastructure and tools required for a comprehensive, multicloud platform that acts as a single point of control.