Page 284 - Data Science Algorithms in a Week

Agent-Based Modeling Simulation and Its Application to Ecommerce 265

The users are modeled as agents with individual behaviors. Risk is modeled into
agent by utilizing the dti, credit history, fico range, income to generate a corresponding
interest rate. Depending on the user state, transitions are triggered by timeouts or by
meeting certain conditions. On executing the program, new borrowers are created who
transition to the PotentialBorrower state. In this state, FICO, DTI and Amount requested
are passed to the neural network class in order to generate a decision on which borrower
transitions to the Screened state. The time spent in a given state follows a uniform
distribution reflecting the time range associated with its state. For example, a typical
Lender takes about 45 days between entry and receiving of first payment. Similarly, the
time spent in the PotentialBorrower state before screening ranges from 2 to 4 days. The
statechart representing borrower and lender behaviors and interactions with the system is
given in Figure 3 and Figure 4.

Figure 3. Borrower statechart.