IncidentTicketAnalysis

Analyze the duration of Incident tickets

Installation

Python: matplotlib.pyplot sklearn.linear_model sklearn.model_selection sklearn.metrics scipy seaborn

Project motivation

https://medium.com/@bartleplae/identify-factors-that-influence-incident-ticket-resolution-timelines-through-python-8a3022b288c4

Business Questions:

1. What are the factors that influence the time to resolve incident tickets?
2. Is there a difference in ticket resolution time between vendors?
3. Can customer satisfaction be related to the resolution time?
4. Does ticket reassignment between groups and staff members affect the resolution time?

While incident ticket resolution timelines are generally available through various dashboards, these timelines can not always be compared given the difference in context such as priority and agreed service level. Therefore looked for a way to standardize the resolution timelines in a way to enable comparisons.

File Descriptions

Code:

• IncidentTicketAnalysis.ipynb (Jupyter notebook / Python code)

Input:

• Incidents.xlsx (subset of fields extracted from EDL through a SQL query and manually anonymized)

Output:

• IncidentDuration_coef.xlsx (coefficients for the factors that influence the time to resolve incidents as determined through the Linear model)
• Factors influencing TTR.xlsx (2 tabs: one for the increases, one for the decreases; shows values for all combination of categorical field / values: mean, t and p value)
• days_to_resolve.png (histogram of the actual time to resolve)
• time_to_resolve_actual_minus_expected.png (histogram of the time difference: actual - predicted)
• assignment_group_company_time_to_resolve.png (boxplot of the time difference, showing the different vendors)
• customer_satisfaction_time_to_resolve.png (boxplot of the time difference, showing the different customer satisfaction scores)
• ticket_reassignment_time_to_resolve (boxplot of the time difference, showing the different reassignment counts)

Technical details

• Step 1: Predict the resolution times based on a set of factors using sklearn linear model
• Step 2: Calculate the difference between the actual and predicted resolution time (this serves as a standardized set which is independent of the known factors)
• Step 3: For each of the remaining factors - variable combinations: determine if the 'mean' of the difference is different from 0 (by means of the 1 Sample T test)

First attempted to use all of the factors as input to the linear model. This resulted in significant overfitting and very high coefficient values. Therefore opted for a 2 step approach using a combination of a linear model and Ttest statistics.

Licensing, Authors, Acknowledgements

• Author:Bart Leplae
• Acknowledgement: Made use of Udacity course materials