Business organizations have been investing heavily in big data and relationship analytics tools to comprehend their customers and competitors in deeper ways. Reporting infrastructure, dashboards, advanced visualizations and other enterprise technologies are helping business managers envisage and use their data efficiently for making smarter decisions.
The importance of analytics is clear but an analytical model must also satisfy some conditions as per the concerned application domain. The first factor here is business relevance which implies that the model should actually resolve the issue for which it was custom developed and not sidetrack from the original problem statement in any phase.
Statistical performance is another important factor. The analytical model should have the statistical power to make predictions. How this is measured will depend upon the kind of analytics considered.
As an example, in a clustering setting the clusters should be as homogeneous as practicable.
According to the nature of business application, the corresponding analytical model should be interpretable and justifiable. With an interpretable model, the user will understand the patterns that it captures. Justifiability refers to the extent to which a model corresponds to previous business knowledge and intuition. Furthermore, it should be remembered that both interpretability and justifiability have to be balanced against statistical performance. It is often seen that high-performing analytical models can be incomprehensible and ambiguous in nature. Conversely, linear regression models are quite transparent and understandable even if they provide only limited modeling power.
The next attribute important for analytical models is operational efficiency. It includes the efforts put in to accumulate data, pre-process it, evaluate the model and feed the results to the application. In a real time online scoring environment, this can be a particularly critical feature. Operational efficiency also concerns the efforts that go into the monitoring and back-testing of the model and re-estimating it when necessary.
Businesses also need to consider the economic cost needed to set up the analytical model. It involves the cost to collect and pre-process the data, the cost to analyze it and the cost to put the consequent analytical models into production. Besides the software costs, the costs of employing human resources and computing resources also need to be taken into account. A detailed cost benefit analysis should be done before starting with the project.
By objecti vity