Thanks For Your Feedback!

Five plus seven is? (answer as number)

Facebook Comments

Click to open/close fB Comments

Predictive Analytics

Predictive analytics is an area of data mining that deals with extracting information from data and using it to predict trends and behavioral patterns. Often the unknown event of interest is in the future, but predictive analytics can be applied to any type of unknown, whether that is in the past, the present or the future.

Predictive analytics uses many techniques from data mining to analyze current data to make predictions about the future, including statistics, modeling, machine learning, and artificial intelligence. For example, logistic regression can be used to turn a market basket analysis into a predictor so that a casino can understand what items are usually purchased together. Of course, the old beer and diapers story market basket wouldn’t fit for a casino, but gleaning data from the casino floor could reveal second favorite games that patrons like to play. This could be useful information when a patron is having a run of bad luck on his or her favorite game. Perhaps a marketing offer for a game he or she sometimes plays would be appreciated rather than an offer on his or her favorite game, as that might not be seen in such a positive light while the patron is in the midst of a losing run.  

Forecasting-Time series regression

Time series regression is a statistical method for predicting a future response based on the response history (known as autoregressive dynamics) and the transfer of dynamics from relevant predictors. Time series regression can help a business understand and predict the behavior of dynamic systems from experimental or observational data. Time series regression is commonly used for modeling and forecasting of economic, financial, and biological systems.

Forecasting-Time series regression work that we have done in the past includes:

  • Time Series models – delivered for The Department of Health and Ageing (Australia) for forecasting national script counts as part of a major project within the Department. Nearly 100 different models were built at a drug type level that could forecast script counts five years in advance. Data set used was 10 years of national script data and these contained over 1.8 billion records. The results revolutionized the Department’s approach to negotiations regarding the PBS scheme. Anecdotal evidence suggests the savings alone from the forecast models and new arrangement were well over AUD $2 billion and allowed in part for current health reform to take place with minimal costs to the taxpayer.
  • Time Series Regression – models were built for the Australian public hospital activity, as part of the move to Activity Based Funding under the National Health Reform program. Separate models were needed for admitted patients, non-admitted patients and outpatients. The Department also required analysis on the effect of environmental changes on hospital utilisation within the Health system for scenario modeling. Such effects included average length of stay, hospital size (staff and beds) and population characteristics. Initial modeling and results helped set the initial price for each and every separation/procedure/operation.
  • Forecasting – a major electronics retailer required models, including exogenous variables that would forecast SKUs at the store level. The nature of the electronics business, and the short shelflife of most products presented many and varied challenges. Final models helped with improvements to the logistic and supply chain process as stock could be better managed. The inclusion of exogenous variables also provided category managers with previously unavailable insight around price elasticity, marketing effectiveness and calendar effects.
  • Time Series Regression Model constructed for a major Australian retailer of fast moving consumer goods, these models revealed the price elasticity of multiple SKUs. Resultant output enabled the retailer to understand the effect of price increases and decreases on sales volume.
  • Commodity Forecasting Models – developed for a multinational construction company, these models gave the business better pricing power. Due to the nature of the construction tender process, multiple years elapse between initial tender submission and the first till of the soil. A lack of understanding of commodity prices saw the companies strikerate for correctly pricing tenders linger around the one-in-two mark, an almost unacceptably low hit-rate. Steel, glass, cement, sand and timber were all investigated and crossed with multiple economic indicators to derive robust forward estimates of prices for a 24-month window.

Multivariate methods

Multivariate statistics is a subdivision of statistics encompassing the simultaneous observation and analysis of more than one outcome variable. The application of multivariate statistics is multivariate analysis.

Multivariate statistics concerns understanding the different aims and background of each of the different forms of multivariate analysis, and how they relate to each other. The practical implementation of multivariate statistics to a particular problem may involve several types of univariate and multivariate analyses in order to understand the relationships between variables and their relevance to the actual problem being studied.

In addition, multivariate statistics is concerned with multivariate probability distributions, in terms of both

  • how these can be used to represent the distributions of observed data;
  • how they can be used as part of statistical inference, particularly where several different quantities are of interest to the same analysis.

Certain types of problem involving multivariate data, for example simple linear regression and multiple regression, are not usually considered as special cases of multivariate statistics because the analysis is dealt with by considering the (univariate) conditional distribution of a single outcome variable given the other variables.

Multivariate work that we have done in the past include:

  • Marketing Response – prediction techniques used to enhance the likelihood of customers responding to marketing campaigns for one of the most recognised and respected organisations in New Zealand. Previously the company had selected targets at random from their customer database and endured response rates around 0.5%. The prediction model that was derived resulted in response rates around 3.5%, a seven fold increase on the previous approach. This piece of work also included the development of a surrogate model to enable the organization to interpret the complex model that was best performed.
  • Prediction Model – constructed for a major Australian retailer, this predicted the likelihood of a customer to be contestable, i.e. to shop with a major competitor. Rich internal data was available for a percentage of customers and this data set was manipulated to reflect market share before a model was built. The results were applied to the remainder of the customer data base, which allowed the retailer to undertake targeted marketing campaigns at those customers with a higher propensity to be contestable.
  • Patron Prediction – US casino property data utilised to construct a pilot model for predicting the long term likelihood of a patron being valuable based off their first visit metrics.
  • Occupancy – Model developed for major Sydney hotel, which looked at likelihood of hotel occupancy reaching a certain threshold a fixed period of time out from the date in question. This could be used to make strategic decisions around raising/lowering tariffs based on what the likely occupancy of was.
  • Australasian Horse Racing Prediction Model(s) – investigated and constructed prediction models for the purposes of wagering. 20 years of race results for all meetings conducted in Australia and NZ were sourced in order to build prediction models. In excess of 4 million records were available, which required cleansing, matching and validation. Variables were derived from the raw indput data. Data driven markets were produced on a daily basis for every NZ horse race.
  • Match Prediction – a number of individual sports models were developed over the years that predict match outcome, margin and likelihood of victory. Coupled with an appropriate set of rules related to staking. These models are used to identify and exploit inefficiencies in wagering markets.