You are on page 1of 2

1|Page

1) Potential Crisis post deployment

a) Network outage in Real-time Solutions

The ML algorithm can be deployed in real-time solutions. The expectation is to store the real-time data
into a historical database from where this data becomes a source of further training the model later on.
Generally the source of data and the ML model outputs are physically apart. So, the data
communication between the real-time system and historical database can be over a network. In case of a
network outage, the data stops flowing to the historical database causing delays in model calibration.

b) Change of data model in the source

The solution is designed to work with a source data model at the time of creation. In case of a change in
data model, the flow or algorithm may partially or completely stop functioning. For example, in case
few columns which early rejected null values, are now accepting null values. These columns may then
require data cleaning and replacement logic before feeding into ML Algorithm.

c) Change in service level agreement of IT services

The current solution has a ticket priority definition that depends on urgency and impact of the incident.
However, the business organization may decide to change the Service Level Agreement (SLA) to
modify the criteria of Priority 1 and Priority 2 tickets. Purpose of such decision may be to improve
support. In this case, the ML algorithm will not perform because it doesn’t know the new definitions of
Priority 1 & 2.

A change in Priority definition will be followed by updating the model and data. The model may have
to learn from the scratch in case there is no mathematical relation found between old Priority definitions
and new ones
2|Page

II. Conclusion.

A. Results

1) Confusion Matrix for Random Forest Algorithm

Confusion Matrix is a measure of performance of a machine learning model for classification problems.
It summarizes the count of correct and incorrect predictions that model has made. These predictions are
categorized into 4 different combinations of predicted and actual values:

 True Positive: Correctly predicted as positive.


 False Positive: Incorrectly predicted as positive.
 False Negative: Incorrectly predicted as negative.
 True Negative: Correctly predicted as negative.

There are the terms which describe the confusion matrix:


 Recall: True Positives divided by total actual positives
 Precision: True positives divided by total positives predicted.
 Accuracy: Out of all classes (both positive and negative), how much were predicted correctly.
 F1-score: Helps in measuring Recall and precision at the same time by using harmonic mean to
suppress extreme values.

You might also like