Sprint Success Forecasting with Machine Learning and Scrum Data - A Machine Learning Model for Sprint Planning

Aug 4, 2026

Upload by Aurelia

Scrum sprints aim to deliver a usable Increment toward a Product Goal, yet Sprint outcomes remain unpredictable and traditional planning aids often perform inconsistently. This master’s thesis studies whether a predictive model trained on Scrum execution data can forecast Sprint success and support more data-driven decisions. Using Jira and Planmill data, a CatBoost gradient boosting model achieves high accuracy, and SHAP analysis reveals influential feature patterns. The model’s practical limits are noted, while it offers value for detecting problematic planning tendencies.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages65

Reading Time2 h 44 min

AccessFree to download

1 Introduction

1.1 Background and motivation of the study

1.2 Research questions

1.3 Scope and objectives

1.4 Thesis structure

2 Agile development and Scrum

2.1 Overview of Agile software development

2.2 Scrum Framework

2.3 Challenges in Sprint predictability

3 Related Work

4 Machine Learning

4.1 Brief history of machine learning

4.2 Machine learning in practice

4.3 Types of machine learning explained

4.4 Problems and ethics in data validity

4.5 Machine Learning Techniques Used

5 Project Background

5.1 Problem statement

5.2 Goals of the project

5.3 Expected benefits of using ML in predicting Sprint outcomes

6 Data Collection and Engineering

6.1 Data collection

6.2 Data preprocessing steps

6.3 Multiple Sprints column

6.4 Exploratory data analysis

6.5 How “SprintSucceeded” was calculated and defined

6.6 Final data cleaning and feature selection

7 Model Development and Training

7.1 Creating and training the first model

7.2 Feature importance analysis

7.3 Creating and training the second model

7.4 Final model with binned values

8 Results and Evaluation

SPRINT SUCCESS FORECASTING WITH MACHINE LEARNING AND SCRUM DATA A Machine Learning Model for Sprint Planning Helmi Harju Master’s Thesis Fall 2025 Degree Programme in Data Analytics and Project Management Oulu University of Applied Sciences ABSTRACT Oulu University of Applied Sciences Degree Programme in Data Analytics and Project Management Author: Helmi Harju Title of thesis: Sprint Success Forecasting with Machine Learning and Scrum Data Supervisor: Ilpo Virtanen Term and year of completion: Fall 2025 Pages: 60 (65) Scrum is one of the most widely used Agile frameworks, where teams work in short, time-boxed iterations called Sprints. Each Sprint aims to deliver a usable Increment that brings the team closer to the overall Product Goal. A common problem while using the Scrum Framework is the unpredictability of Sprint outcomes. Traditional decision-making tools are often used in Sprint planning, with varying degrees of success. This thesis aimed to explore whether a model trained on Scrum data could predict Sprint outcomes, specifically their successes, and enable Scrum teams to make more data-driven decisions based on these predictions. This thesis also aimed to explore the features of Scrum data that were most influential to the outcome of the Sprint. Data from Jira and Planmill were used to train a model with CatBoost, a Gradient boosting decision tree algorithm. The resulting model had a high accuracy rating, and a SHAP analysis highlighted features that aligned with real-world factors. Although the model’s predictive power is limited, it demonstrates potential as an analytical tool for identifying problematic Sprint planning tendencies. While researchers have previously trained models that can assign Story points or man-hours based on Scrum data, the topic of this thesis has not been explored before. This thesis serves as a starting point for a completely new research topic. Using a bigger dataset or exploring other machine learning techniques are recommended future improvements. CONTENTS ABSTRACT ......................................................................................................... 2 CONTENTS ........................................................................................................ 3 1 INTRODUCTION.......................................................................................... 5 1.1 Background and motivation of the study............................................. 5 1.2 Research questions ............................................................................ 6 1.3 Scope and objectives .......................................................................... 6 1.4 Thesis structure .................................................................................. 7 2 AGILE DEVELOPMENT AND SCRUM........................................................ 8 2.1 Overview of Agile software development............................................ 8 2.2 Scrum Framework .............................................................................. 9 2.2.1 Introduction of Scrum History................................................ 10 2.2.2 Scrum Team overview .......................................................... 11 2.2.3 Scrum events ........................................................................ 13 2.2.4 Scrum artifacts ...................................................................... 16 2.3 Challenges in Sprint predictability..................................................... 16 3 RELATED WORK ...................................................................................... 19 4 MACHINE LEARNING ............................................................................... 21 4.1 Brief history of machine learning ...................................................... 21 4.2 Machine learning in practice ............................................................. 22 4.3 Types of m

FAQs

How does the thesis define and predict Sprint success?

It trains a model to forecast Sprint outcomes, specifically whether a Sprint is successful, using engineered data features derived from Scrum execution records.

What data sources and machine learning approach are used?

The model is trained using data from Jira and Planmill, employing CatBoost, a gradient boosting decision tree algorithm.

How are the most influential features identified?

SHAP analysis is used to highlight features whose effects align with real-world factors affecting Sprint outcomes.

Aug 4, 2026

Upload by Aurelia