ITSM Incident Intelligence Project

Project Description

This project uses a synthetic ITSM incident dataset to build predictive and forecasting models that support service operations planning. It focuses on understanding incident behavior, predicting ticket characteristics, and estimating future ticket load.

What The Project Solves

• Predicts incident Priority from ticket metadata.
• Predicts whether an incident is associated with any related change activity.
• Forecasts daily incident volume using time-series modeling.

Data Context

The source data (dat.csv) is incident-level tabular data with categorical, numeric, and timestamp fields (for example: CI category, subcategory, WBS, priority, and open time).
For forecasting, the incident records are transformed into a daily count series (No_Incidents) indexed by date.

Modeling Approach

• Classification track Uses reproducible preprocessing + model pipelines for imputation, categorical encoding, and feature scaling where needed.
  Model comparison is driven by imbalance-aware metrics such as macro F1 and balanced accuracy.

• Forecasting track Normalizes mixed date formats, aggregates daily incident counts, and applies SARIMAX for short-horizon forecasting.
  Performance is assessed with both holdout evaluation and rolling-origin backtesting.

Outputs

The project produces model artifacts and evaluation reports that document:

• model selection results and per-target classification performance,
• confusion matrices and detailed classification metrics,
• forecast quality metrics and benchmark comparison against a naive baseline.

Repository Purpose

This repository is intended as an end-to-end analytics and modeling reference for ITSM incident intelligence, combining classification and time-series forecasting in a single workflow.