EFFORT ESTIMATION IN AGILE PROJECTS USING ADAPTIVE AI-DRIVEN APPROACH
Keywords:
Agile Software Development, Effort Estimation, Adaptive Estimation Framework, Machine Learning in Agile, PID Controller, Real-Time Feedback, Estimation Accuracy, Sprint Analytics, Human-Centered AI, Predictive Modeling, Software Project Management, Closed-Loop Control SystemAbstract
Changing team dynamics, faulty historical data, and constantly evolving project requirements make it difficult to estimate effort in Agile software development. To address these challenges, we're putting forth a novel strategy dubbed the Adaptive Effort Estimation Approach (AEEA), which blends Machine Learning (ML) with Proportional-Integral-Derivative (PID) control techniques. AEEA continuously changes over time, much like a feedback system, in contrast to conventional static ML models. We used real-world and simulated data from different Agile teams to test this framework. The findings demonstrated that, in contrast to conventional techniques like expert judgment and Planning Poker, AEEA improved estimation accuracy and reliability.
Rapid iterations and shifting requirements create inherent uncertainty in Agile systems. These ambiguities may result from a variety of risk factors, including:
- Incomplete or vague user stories
- Unexpected team member absences or turnover
- Integration issues or third-party service delays
- Unforeseen technical complexity
- Unplanned blockers (e.g., missing dependencies, critical bugs)
These hazards have a direct effect on effort estimation because they might cause planned workloads to be inflated or disrupted. These erratic risk events are not dynamically taken into account by conventional estimation methods, whether they be expert-driven or story point-based.
This research incorporates AI-driven risk forecasting, which:
- Predicts potential risks before the sprint starts using historical sprint data, team volatility patterns, and prior deviations.
- Feeds this risk insight into the estimation model, allowing effort predictions to reflect a more realistic workload.
- Enhances the accuracy and adaptability of the estimation process.
- The model shifts from static planning to a more context-aware and adaptive system by adding risks as an explicit forecasting input, where:
- High-risk sprints prompt higher buffer or adjusted capacity planning.
Low-risk environments allow for more aggressive sprint targets.
