nuluh/thesis
1
0
Fork 0
Code Issues 47 Pull Requests Actions Packages Projects Releases 1 Wiki Activity

[FEAT] Include Undamaged Node Classification #53

New Issue
Open
opened 2025-05-06 05:06:27 +00:00 by nuluh · 1 comment
nuluh commented 2025-05-06 05:06:27 +00:00 (Migrated from github.com)
Copy Link

Problem Statement

The current machine learning model is only trained to identify and classify damaged nodes among different damage locations. However, it lacks the capability to properly classify undamaged nodes, which is a significant limitation. When the model is presented with data from an undamaged structure, it will incorrectly classify it as one of the damaged categories since it has no concept of an "undamaged" state.

Proposed Solution

Expand the machine learning model to include undamaged node classification by:

  1. Including undamaged (healthy) structure data in the training dataset
  2. Adding "undamaged" as a distinct class label in the classification scheme
  3. Modifying the training pipeline to incorporate this new class
  4. Retraining the model with the expanded dataset and label set
  5. Evaluating the model's performance on both damaged and undamaged classifications

Alternatives Considered

  1. Two-stage classifier: First classify damaged vs. undamaged, then identify damage location if damaged. This adds complexity but might improve specificity.
  2. Anomaly detection approach: Train on only undamaged data and detect deviations. This would change the fundamental approach from classification to anomaly detection.
  3. Confidence threshold: Keep current model but establish a confidence threshold below which predictions are considered "undamaged." This is simpler but less reliable.

Component

Python Source Code

Priority

Critical (blocks progress)

Implementation Ideas

No response

Expected Benefits

  1. Comprehensive classification: Model will correctly identify both damaged and undamaged structures
  2. Reduced false positives: Fewer healthy structures will be incorrectly classified as damaged
  3. Complete SHM system: More practical for real-world structural health monitoring applications
  4. More robust evaluation: Performance metrics will reflect real-world usage scenarios
  5. Better thesis contribution: Demonstrates a complete solution rather than partial capability

Additional Context

This enhancement addresses a fundamental limitation in the current model. For a structural health monitoring thesis, the ability to distinguish between damaged and undamaged states is essential for practical application. The confusion matrix and performance metrics should be updated to reflect this expanded classification task.
Consider creating a new experiment notebook specifically focused on evaluating the model's performance in distinguishing undamaged structures from various damage locations. This will provide valuable insights into the model's practical utility.

Key metrics to evaluate after implementation:

  • Overall accuracy
  • False positive rate (undamaged classified as damaged)
  • False negative rate (damaged classified as undamaged)
  • Confusion matrix including undamaged class
  • Class-specific F1 scores
### Problem Statement The current machine learning model is only trained to identify and classify damaged nodes among different damage locations. However, it lacks the capability to properly classify undamaged nodes, which is a significant limitation. When the model is presented with data from an undamaged structure, it will incorrectly classify it as one of the damaged categories since it has no concept of an "undamaged" state. ### Proposed Solution Expand the machine learning model to include undamaged node classification by: 1. Including undamaged (healthy) structure data in the training dataset 2. Adding "undamaged" as a distinct class label in the classification scheme 3. Modifying the training pipeline to incorporate this new class 4. Retraining the model with the expanded dataset and label set 5. Evaluating the model's performance on both damaged and undamaged classifications ### Alternatives Considered 1. Two-stage classifier: First classify damaged vs. undamaged, then identify damage location if damaged. This adds complexity but might improve specificity. 2. Anomaly detection approach: Train on only undamaged data and detect deviations. This would change the fundamental approach from classification to anomaly detection. 3. Confidence threshold: Keep current model but establish a confidence threshold below which predictions are considered "undamaged." This is simpler but less reliable. ### Component Python Source Code ### Priority Critical (blocks progress) ### Implementation Ideas _No response_ ### Expected Benefits 1. Comprehensive classification: Model will correctly identify both damaged and undamaged structures 2. Reduced false positives: Fewer healthy structures will be incorrectly classified as damaged 3. Complete SHM system: More practical for real-world structural health monitoring applications 4. More robust evaluation: Performance metrics will reflect real-world usage scenarios 5. Better thesis contribution: Demonstrates a complete solution rather than partial capability ### Additional Context This enhancement addresses a fundamental limitation in the current model. For a structural health monitoring thesis, the ability to distinguish between damaged and undamaged states is essential for practical application. The confusion matrix and performance metrics should be updated to reflect this expanded classification task. Consider creating a new experiment notebook specifically focused on evaluating the model's performance in distinguishing undamaged structures from various damage locations. This will provide valuable insights into the model's practical utility. Key metrics to evaluate after implementation: - Overall accuracy - False positive rate (undamaged classified as damaged) - False negative rate (damaged classified as undamaged) - Confusion matrix including undamaged class - Class-specific F1 scores
nuluh commented 2025-05-06 05:13:38 +00:00 (Migrated from github.com)
Copy Link

Should I only use one file (zzzAU.TXT) or use all the file based on column sensors damaged which used and which being leaved out as undamaged data?

Should I only use one file (zzzAU.TXT) or use all the file based on column sensors damaged which used and which being leaved out as undamaged data?
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
bug
Something isn't working
 critical
Block Progress
 documentation
Improvements or additions to documentation
 documentclass
Improvements or additions to LaTeX documentclass
 duplicate
This issue or pull request already exists
 enhancement
New feature or request
 good first issue
Good for newcomers
 help wanted
Extra attention is needed
 invalid
This doesn't seem right
 latex
Related to the writings
 question
Further information is requested
 wontfix
This will not be worked on
No Label critical enhancement
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
nuluh
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: nuluh/thesis#53
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?