Workforce optimization for enhancing allocation quality of bookings in domiciliary care

• Main Author: Bakshi, Ayushi
• Other Authors: Thambipillai Srikanthan
• Format: Thesis-Master by Research
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Engineering
• Online Access: https://hdl.handle.net/10356/181290
• Institution: Nanyang Technological University

Workforce planning and optimization in the domiciliary care industry represent intricate, yet essential endeavours filled with challenges. The predominantly manual nature of these efforts leads to suboptimal outcomes that are also prone to errors. Coupled with hiring decisions that are often made without a clear understanding of the requirements, this further contributes to care worker retention issues due to unfair workload allocations. This necessitates a systematic approach to workforce planning and optimization in order to improve the satisfaction levels of both care workers and clients. In this thesis, novel techniques have been proposed to improve allocation quality of booking, thereby enhancing the service quality in a domiciliary care setting. First, techniques for the systematic identification of any gaps in the dataset were proposed and validated. The proposed approach relies on an auto-allocation process to highlight critical gaps in the dataset such as the absence of minimum number of care workers to satisfy the gender restrictions, skills required to serve the bookings, unavailability of care workers for bookings, etc. This has ensured that the proposed techniques for allocation quality improvement are adaptable to any dataset. An interactive process was introduced to ascertain that the permitted relaxation of care worker attributes can yield the user-desired improvement to allocation quality. This approach has shown to be effective in guiding the users to establish a well-informed and attainable objective prior to the identification of feasible solutions. Experimentation based on a data set consisting of 700 bookings show that the proposed techniques for dataset validation can be completed in less than 3 minutes to ensure that the subsequent detailed analysis can yield the desired outcomes demanded by the user. Next, techniques for optimizing the effectiveness of existing care workers to improve the allocation quality were proposed. This involved the systematic adjustments to care worker attributes, such as skills, nominal hours, availability and preferences so as to minimize disruptions to care workers and operations without compromising on the improvements to allocation quality. The proposed technique relies on establishing the maximum possible improvement to allocation quality prior to invoking a systematic customization process for maximizing the allocation quality improvement. Priority is given to unallocated and lowest-quality bookings in an attempt to maximize the effectiveness of existing care workers. Moreover, the proposed approach has paved the way for proposing alternate solutions to facilitate the selection of a most appropriate recommendation by the user. Despite the intricate interplay of workforce attributes on allocation quality, the turnaround time for recommending alternate solutions was just under 10 minutes for a dataset of 700 bookings. Following this, techniques for injecting new care workers to the existing supply of care workers were proposed to maximize the allocation quality improvement. This was achieved by targeting the low-quality bookings in a neighbourhood. K-means clustering technique was employed to identify optimal number of clusters for maximizing the utilization of new care workers. A greedy approach has been adopted within each cluster to ensure that the new care workers are assigned to the lowest quality bookings in close proximity for maximizing allocation quality improvement. Moreover, an iterative technique has been incorporated to prioritize the allocations of the lowest quality bookings for maximizing the allocation quality improvement. This coupled with a systematic elimination of ineffective care workers in each cluster have resulted in the most effective placement of new care workers in each cluster. The proposed techniques have shown to be capable of targeting the new care workers to bookings in most need, thereby providing an attractive alternative to existing suboptimal manual efforts. Experiments show that, for a dataset of 700 bookings and new care worker threshold of 5, the turnaround time for recommending the attributes and their respective geolocations for new care workers is about 7 minutes. A framework has been proposed to highlight the systematic integration of the proposed techniques in this thesis. The main sequence consists of dataset validation, optimization of existing care workers and maximizing the effectiveness of available new care workers. Alternative options available for optimizing existing care workers allow for greater flexibility in the assignment of new workers. The framework also supports the rapid identification of suitable care workers for unallocated bookings within the constraints of user specified attribute relaxations. Moreover, an elegant approach has been incorporated for forecasting the necessary changes to the care worker supply to support new care plans. The proposed techniques provide for a systematic approach to facilitate the gap analysis and forecasting for domiciliary care organizations. Finally, the proposed techniques have been validated with realistic dataset to demonstrate their effectiveness and efficient runtimes, making them highly suitable for real-life deployment.