IQRiscoman

Scoring, classification and recommendation systems are valuable to use for supporting decisions and to identify best courses of actions. We provide methods/ systems to organize priorities, define multi criteria and multi objectives for organizations properly perform. We go further in the creations of the scores / classification and recommendation systems analyzing the risk component of the possible results.

The problem of risk classification is one of the most important topics in banking, insurance, reinsurance, financial industry and health services. It involves both underwriting and the actuarial calculation of prices, reserves, analysis of ruin problems, diagnostic, follow-up actions, etc.. The process requires identifying classes (coming in principle from unsupervised learning) such as good or bad prospects, the probabilities of monetary losses on a portfolio, and the survival of loans until maturity. Risk classification also has a probability of misclassification that should be minimized. However, there is a priority problem in risk classification: how to define the classes or groups that are required to separate risks and then to find membership to the classes of new profiles that are in the system.

In addition, risk classification requires taking into account some very important elements of analysis. Risk can be classified according to frequency, such as good and bad drivers in different time periods, regions, or other variables. The intensity of the risk can also vary, and it is possible to have a small number of large claims. The company’s strategy can be based on a trade-off between the treatment and expectation of risks frequency and intensity. This means that there are different groups of variables to make decisions in terms of a model.

Classification methods require complete data, but sometimes the data is non-existent or incomplete. Some variables are difficult to find in a file. For example, high-value vehicle parts have a large impact on the value of a claim, but data availability on them is not easy. Loss distributions are generally long-tailed, far from normal conditions, and sometimes mixed distributions. The effect of time is very important; a risk may be good today but may change tomorrow due to external factors. The vast majority of variables in insurance are categorical, and it is necessary to create models that use them properly.