Advanced Analytics

expertise advancedanaylitics

EurekaFacts has an extensive experience conducting data mining, multivariate analyses, statistical, econometric, and predictive modeling for private firms, non-profit organizations, and Federal agencies. We master both supervised and unsupervised machine learning. The most common techniques we have used include:

  • For a federal agency, EurekaFacts used regression analysis to model weekly web-traffic to one of its websites as a function of different ad campaigns on local and national TVs, and identified the impact of the individual ad campaigns on the target audience. The statistical modeling included a time-series analysis to control for the “natural” web-traffic variation and its seasonality using historical data. The model identified ads with the overall greatest impact on the target population, as well as their differential impacts at local level.
  • For a private company, we used Bayesian analysis to incorporate findings from a previous research effort into their new marketing strategy. Bayesian analysis is particularly useful as it allows to take advantage of prior knowledge of the phenomenon we’re studying.
  • For a financial institution, we used logistic regression to identify customers more likely to engage in a given financial transaction. Logistic regression is used to estimate the likelihood of a dichotomous outcome (success/failure, presence/absence, yes/no) based on a set of predictors.
  • For a State Transportation Administration, we used a multinomial logit to model motorists’ likelihood to choose a toll lane to save time over toll-free highway lanes, considering different travel scenarios. Multinomial logit is used to model discrete choice outcome with a set of predictors.
  • Decision trees are tree-like diagrams used to show the variations of the outcome for different values of predictors in an “if-else” way.
  • For a wireless carrier, we applied neural networks to community-level socio-demographics characteristics and store sales to identify the most profitable locations for new stores. Neural networks use different combinations of a set of predictors to predict one or more outcomes.
  • For a financial institution, we used cluster analysis for market segmentation and customer insights to support competitive strategy. Cluster analysis is used to identify natural groupings of individuals described by a common set of parameters. This technique is particularly suitable when prior classification of observations (for example customers) is not available.
  • Similar to cluster analysis, association rules can be used to identify cross-selling opportunities. For a financial institution, we used association rules to tell how likely customers who subscribe to products X and Y are to also subscribe to product Z.
  • For a wireless carrier, we used k-nn to predict the likelihood of a customer to buy a premium service, considering that most customers with similar characteristics bought the premium service. k-nn (k nearest neighbors) is used to classify a new observation (or customer) in the dominant class among its k neighbors in the space defined by the set of predictors.
  • Experimental design is used to investigate hypotheses in a controlled environment where a framework is specified to observe, measure, and evaluate groups in response to a designated response.