Analyzing predictions and postprocessing

Before we charge off to deploy the model, it would be advisable to analyze the predictions and see if they make sense, whether there are some patterns in the errors, and also how to turn the predictions into something actionable. These are aspects where traditional data science tools and methods are not of much help, and you need to rely on judgment and methods from other disciplines to help formulate the next steps. For this, let's start by combining the scoring dataset file with the explanations file. This can be done in Structured Query Language (SQL), Python, or Excel. The combined file looks something like this:

Figure 8.6 – Combined scoring data and predictions

We also created a new ERROR column that simply subtracts prediction from price. We can now use Excel to create a pivot table and look at the results from multiple perspectives. For example, let's create a pivot table and look at the Average...