First, we would need to have an application; let's go ahead and create an Angular application as shown previously.
From the previous example, we have seen that the first thing we will need is to list out our training data, which will be fed into our algorithm. In this case, we will need to first identify the different attributes that are affecting our target attribute (approved). Without going in too deep, the following are the major factors (and their possible values), which we are taking as an example of something, which can affect your approval chances:
- Credit score: Overall score of your credit (Excellent, Good, Average, and Poor)
- Credit age: Age of your credit history in years ( >10, >5, >2, >=1)
- Derogatory remarks: If there are any remarks on your account (0, 1, 2, >=3)
- Utilization: How much of your approved credit you use (High, Medium, and Low)
- Hard inquiries: How many new accounts did you open recently (0, 1, 2, >=3)
Because of the fixed number of combinations of the preceding list, we can, in theory, generate a dataset that comprises all scenarios and then we can predict with 100% accuracy using that dataset, but where is the fun in that. We will instead be only taking half of the generated dataset and use it to predict the outcome of the other half.