Root Cause Prediction Using Deep Learning

Problem Statement

In IT Service Management (ITSM), incident tickets often provide user-reported symptoms. Identifying the root cause of these symptoms is crucial for reducing resolution times and improving user satisfaction. This project employs deep learning to predict the root cause based on the given symptoms.

Key Objective: Quickly identify the root cause of incidents reported in ITSM based on observed symptoms.

Dataset

Dataset Link: Root Cause Analysis Dataset on Kaggle

Overview

The training data consists of symptoms collected by technical teams during incident analysis. Each symptom is represented as a binary attribute (1 or 0), indicating its presence or absence. The dataset includes 1000 training examples, providing a foundation for predicting root causes.

Symptoms Include:

• CPU load
• Memory load
• Network delays
• Specific error codes in log files

Target Variable:

• Root Cause (Classes of root causes)

Note: In real-world scenarios, symptom diversity may be vast. Symptoms can be further categorized by location (e.g., load balancer, microservice) and isolated from text using text mining.

Model Architecture:

Model Overview

The neural network architecture consists of hidden layers using the relu activation function and an output layer using softmax. The model is trained with categorical cross-entropy loss and the Adam optimizer.

Activation Functions

• Hidden Layers: relu
• Output Layer: softmax

Loss Function

• Categorical Cross-Entropy

Optimizer

• Adam

Model Performance

• Training Accuracy: 87%
• Validation Accuracy: 81%

Deployment

The trained model can be deployed for practical use. Below is a screenshot of the web application: