callforecast Hackathon

Sponsored by Weber State University and Clear View Live Data Analytics

Originally developed for a statistical modeling competition, which placed 3rd overall. This repo demonstrates how to forecast call-center volume and average handle time (AHT) in 30-minute intervals using machine-learning and statistical methods.

Table of Contents

1. Project Overview
2. Data

• 3. TODO Modeling Approach
• 4. TODO Results

---

Project Overview

Call centers require accurate predictions of:

1. Call Volume (CV): How many inbound calls arrive in each 30-minute interval.
2. Average Handle Time (AHT): The average agent time spent on each call.

By forecasting these metrics, companies can:

• Optimize staffing schedules to reduce costs and wait times.
• Identify performance trends to guide agent training.
• Improve infrastructure planning and reduce idle agent time.

This codebase implements multiple models (including linear regression, GAM, and gradient boosting) to predict call volume and AHT using historical call-center data.

---

Data

The dataset (training.csv and test.csv) contains:

• Timestamps of each call (Call Start)
• The length of time an agent spent on each call (Agent seconds)
• Other potential features like day of week, time of day, etc.

Data coverage:

• TODO

Note: For AHT, calls with 0 agent seconds were excluded from modeling (as these likely represent hangups before agent interaction).

---