Scotiabank OEC Data Simulation

Overview

This repository contains code to run a discrete-event simulation that creates synthetic event logs for Scotiabanks OEC system for testing and demonstration purposes. It models how employees create and work on customer complaints, requests, and claims.

Setup

To setup the simulation on your own machine run the following commands in order:

git clone https://github.com/QuMuLab/scotiabank-data-simulation.git

cd scotiabank-data-simulation

python -m venv oec_simulation

source oec_simulation/bin/activate # macOS or Linux

oec_simulation\Scripts\activate # Windows

pip install -r requirements.txt

Usage

The simulation can be ran in headless mode with:

python simulation/OEC_simulation.py

Or can be ran through an interactive dashboard with:

python simulation/OEC_simulation.py --dashboard

Configuration

In headless mode, the simulation settings will pull from the values in config/preferences.yaml, you can edit this file as you see fit and re-run the simulation to adjust it's output. Alternatively, you can pass a flag while running in headless mode with the name of a parameter and a value to overwrite that parameter's value in the yaml file.

For example:

python simulation/OEC_simulation.py --sla_multiplier 0.2 --base_reassign_chance 0.4

Run python simulation/OEC_simulation.py --help for a full list of the accepted parameters. But note that parameter flags will not work if the script is run with --dashboard.

If you ran the simulation with --dashboard you can edit the settings through the dashboard, which is accesible at http://127.0.0.1:8050. You can easily adjust the settings without having to rerun the script by clicking "Back to Setup" after a run of the simulation. The dashboard autofills with the defaults from preferences.yaml.

Changes made directly to preferences.yaml will save between runs of the simulation, but changes made through the dashbaord or through the command line flags will not.

Additionally, you can change what kind of incidents are created by adding, removing, or editing variants in config/incidents.yaml.

Parameters

A brief description of what each parameter in preferences.yaml does can be found below.

settings

Key	Description
start_date	Simulated calendar start date
simulation_days	Number of days the simulation runs for
work_day.start / work_day.end	The business hours employees will work
employee_count	Number of employees
transaction_date_window	Max number of days old an incident's transaction day can be
employee_timeout.minimum / .maximum	Hours an idle employee waits before checking for new work
employee_work_session.minimum / .maximum	Hours a single work session on an incident can be
employee_work_break	Hours max between work sessions
employee_clarification_wait.minimum / .maximum	Hours to wait on a customer clarification response
noise_percentage	Gaussian noise applied to the probabilities & weights each run

probabilities

Key	Description
select_unassigned_incident_chance	Chance an employee pulls from the unassigned queue instead of creating a new incident
select_new_incident_chance	Chance a newly created incident is worked on immediately rather than queued
sla_multiplier	Multiplier applied to each incident variant's base SLA
incident_high_priority_threshold	Fraction of SLA time remaining for an incident to be "high priority"
hand_over_chance	Chance an incident is handed off to another employee after a break
base_clarification_chance	Base chance of requesting a customer clarification
incident_resolution_sessions_weight	Added resolution chance per completed work session
incorrect_field_chance	Base chance a given field is populated incorrectly
fraud_clarify_multiplier	Multiplier on clarification chance for suspected-fraud incidents
clarification_retry_chance	Chance of a repeat clarification when fixing an incorrect field
clarification_retry_decay	Decay applied to retry chance on each subsequent attempt
base_reassign_chance	Chance an incident is reassigned to a different transit
base_cancel_chance	Base chance an incident is cancelled instead of closed
incident_blowup_chance	Chance of an access-control "burst" event on an incident

weights

Key	Description
providers	Relative weights for provider values (e.g. TSYS, NOT_APPLICABLE, FDR) assigned to Claims/Requests incidents
priorities	Relative weights for High / Medium / Low incident priority
reception_channels	Relative weights for how an incident was received (Branch, Phone Call, Email, etc.)
missing_fields	Relative weight controlling how often each listed field is generated incorrectly or missing
root_causes	Per incident type (Claims, Requests, Complaints), relative weights over incident root causes
clarification_reasons	Relative weights over the reason given when a clarification is requested
factor_decays	Decay factors (Sessions, Clarifications) used when scoring an incident's closure response type