In this chapter, the authors show how to build a decision tree from given real-time data. They interpret the output of decision tree by learning decision tree classifier using really recursive greedy algorithm. Feature selection is made based on classification error using the algorithm called feature split selection algorithm (FSSA), with all different possible stopping conditions for splitting. The authors perform prediction with decision trees using decision tree prediction algorithm (DTPA), followed by multiclass predictions and their probabilities. Finally, they perform splitting procedure on real continuous value input using threshold split selection algorithm (TSSA).
TopIntroduction
Readers want to understand the pitfalls in handling real time financial data, in which Decision trees are extremely useful in representing the real time scenario of finance industry. Our aim in this chapter is to explore that domain, in particular during evaluating loan applications. Consider, a scenario where an individual needs to buy a house, to which they may not have all the money, where there is a need to take a loan from the bank. The bank is going to look at some properties of one's history, like:
- 1.
Credit
- 2.
What has it been like in the past?
- 3.
How much money does He/She actually make?
- 4.
How long His/Her loan is off?
- 5.
What time He/She is willing to spend before paying it off ?
- 6.
Personal information about him, like gender, age and so on.
The bank will collect the information and try to make a prediction, seeing if sanctioning a loan to the individual is a risky factor or not. Typically, Industry experts look at credit history, which looks at all the other loans taken in the past and credit cards and all that and have paid those off on time.
This chapter is organized as follows: First, we are learning how to build a decision tree from data, followed by Interpreting the output of decision tree. Next, we are learning decision tree classifiers using really recursive greedy algorithm, followed by selection of the best feature to split, based on classification error called feature split selection algorithm, with all possible stopping condition for splitting. Next, we are learning how to perform predictions with decision trees using decision tree prediction algorithm, followed by multiclass predictions and their probabilities and finally, performing splits on real continuous value input using threshold split selection algorithm.
TopLiterature Review
In a direction to understand the amount of research carried in applying the decision tree in predicting the loan risk factor, Mistikoglu et al., (2015) adopted C5.0 and CHAID algorithms to construct decision trees and to extract rules that show the associations between the input and output variables for roofer fall accidents. Important input variables of this models were identified as the fall distance, injury cause, safety training, and construction operation prompting fall, considering that these factors had the best predictive value related to whether a roofer fall accident can result in a fatality (or) nonfatal injury. Bahnsen et al. (2015) proposed an example dependent cost sensitive decision tree algorithm, by incorporating the different example dependent costs into a new cost based impurity measure and a new cost based pruning criteria. The study used three different databases, from three real-world applications as credit card fraud detection, credit scoring, and direct marketing, and evaluated that the proposed algorithm is the best performing method for all databases by comparing against a standard decision tree. Kabir et al. (2017) made a review on the standard fault tree with different extensions of standard fault trees and prominent Model Based Dependability techniques, in which fault trees are used as a means for system dependability analysis and provide an insight into their working mechanism, applicability, strengths and challenges. Whereas Fletcher et al. (2017) proposes a new differentially-private decision forest algorithm that minimizes both the number of queries required, and the sensitivity of those queries and build an ensemble of random decision trees that avoids querying the private data except to find the majority class label in the leaf nodes, uses the Exponential Mechanism to only output the class label itself. The author Xia et al. (2017) highlighted the hyper parameter tuning of base learners despite being critical to well performed ensemble models, built sequential model boosting, focused on developing the same or different algorithms in parallel, and focused on the comprehensibility of models. The author proposed a sequential ensemble credit scoring model based on a variant of gradient boosting machine. The model mainly consists of three steps.