This model is a fine-tuned version of BERT (bert-base-uncased) designed to classify the impact of incident records based on their short descriptions. The impact is categorized into three levels: low, medium, and high.
Intended Use
Only for Demo pupose - The model is intended to assist in the automatic categorization of incident impacts to streamline incident management processes. It can be used by IT service management teams to quickly identify the severity of incidents based on short descriptions.
How to Use
Inference
To use the model for inference, you can utilize the Hugging Face transformers library. Below is an example of how to load the model and make predictions:
python
1from transformers import pipeline
23model_name ="xeroISB/incidentImpactModel"4classifier = pipeline("text-classification", model=model_name)56short_description ="Network outage in building 12"7prediction = classifier(short_description)8print(prediction)
Training
The model was trained using the following configuration:
Model: BERT (bert-base-uncased)
Learning Rate: 2e-5
Batch Size: 16
Epochs: 3
Evaluation Strategy: Epoch
Optimizer: AdamW
Loss Function: Cross-Entropy Loss
Dataset
The dataset used for training includes the following columns:
short_description: A brief description of the incident.
impact: The impact level categorized into three classes: low (3), medium (2), and high (1).
The impact values were mapped to integer labels as follows:
3 (low) -> 0
2 (medium) -> 1
1 (high) -> 2
Tokenization
The short_description was tokenized using the BERT tokenizer with padding to the maximum length and truncation enabled.
Performance
Confusion Matrix
The confusion matrix on the validation set is as follows:
