Business Process Performance Prediction on a Tracked Simulation Model Andrei Solomon, Marin Litoiu– York University
Agenda ›Motivation ›Proposed Architecture ›State Prediction ›Results ›Conclusions 9/3/2015 2
Motivation 9/3/ ›Business processes ›need to adapt to satisfy service level agreements ›monitor ›determine changes ›Execute
Motivation 9/3/ › analyzing the data › quantitative evaluation of different change decisions › process optimization › needs forecasted key performance indicators › to asses the effect of changes › limitations of current approach: › forecasts based on simple interpolation inaccurate predictions and wrong decisions Benefits feedback based evolution architecture that + business agility + more accurate simulation + more accurate predictions + more accurate decisions
States and KPI States: Raw monitoring metrics ▫Individual task durations ▫Message length and frequency ▫Number of users, etc.. KPIs: Example: Average Process Duration KPI KPI definition - specifies the method of calculation, given: ▫current instances ▫aggregated metrics ▫predefined set of aggregation functions (i.e. average) ▫time period for data collection (example: rolling 30 days = 30 days sliding window) ▫specifies a desired target Are defined in Modeling phase
Predictive Feedback Loop ›Goal ›maintain KPIs close to the reference target ›predict short term change ›to enable more effective planning and strategic decisions ›using estimated states 9/3/2015 6
Case Study (Credit Approval) Estimation, prediction and integration: our contribution 9/3/ IBM WebSphere Integration Developer (WID) IBM WebSphere Business Modeller IBM WebSphere Process Server + Monitor
State Prediction – Linear Regression 9/3/2015 8
State Prediction - ARIMA 9/3/2015 9
Without Simulation 9/3/
KPI Prediction - Results ›(a) Err = 23% 9/3/ ›(b) Err = 21% ›(c) Err = 7%
Conclusions & Future work Conclusions: ›feedback based evolution architecture ›automated live monitoring ›a KPI prediction module ›Forecasts the states (linear regression and ARIMA) ›Uses a simulator to correlates the states Further work and future challenges include: ›validation - other estimators ›modeling human resources ›implement an optimization algorithm 9/3/
Thank you. Questions? 9/3/