Our methodology for building predictive analytics solutions is based on an iterative 4-step process.
First, we conduct short interviews with clients to determine the objective of their analysis. Examples of objectives include:
The next step is examining your data. We work with data in any form: text files, Excel worksheets, or SQL databases are not a problem. For a one-time analysis we will usually take the data in whatever form you have available. If the analysis is going to be recurring, we will help you design a process for extracting the data from the source system, checking the data for errors, and transforming the data for use with our analytical tools.
Preparing the data for analysis can be a large part of the project. Factors that influence the amount of data preparation work include the quantity of missing values and the degree to which the data has already been normalized or coded. As an example, consider an analysis involving age. If some of the ages are recorded as intervals like 30 – 39 and others are recorded as whole numbers like 54 then we will have to recode this field in a consistent manner.
The third step in our process is determining the appropriate methods to solve your problem. We use traditional statistical techniques such as least-squares regression, more modern approaches such as multilevel/heirarchical models, and statistical learning approaches such as decision trees or support vector machines. Some of the factors that guide this decision include:
We provide you with the results of the analysis in a written report using appropriate tables and visualizations. We determine to what degree the results are statistically and economically significant and make recommendations for future work.