PREDICTIVE FACTOR ANALYSIS AND USER SEGMENTATION FOR DEVELOPING IMPLEMENTATION APPROACHES FOR A KPI MANAGEMENT SYSTEM IN PRINCE OF SONGKLA UNIVERSITY, PHUKET CAMPUS

Abstract

This study aimed to 1) examine the current state, problem levels, and risks associated with internal employee Key Performance Indicator (KPI) data management; 2) identify critical predictive factors influencing the current state and risk levels using Random Forest Regression; and 3) segment system users based on their perceived current state and risk levels using
K-Means Clustering to propose tailored implementation guidelines for each user group.
The sample consisted of 167 personnel from Prince of Songkla University, Phuket Campus, selected via stratified random sampling based on department affiliation. Data were collected using an online 5-point Likert scale questionnaire. The research instrument demonstrated an Item-Objective Congruence (IOC) index of 0.67 and a Cronbach's Alpha reliability coefficient of 0.89. Data were analyzed using descriptive statistics, Random Forest Regression to extract Feature Importance Scores (MAE = 0.75, RMSE = 0.91), and K-Means Clustering (k = 3), which yielded a Silhouette Score of 0.39 and a Davies-Bouldin Index of 0.87, confirming the validity of the three-cluster classification. The findings revealed that the overall current state of KPI data management was at a high level, with the alignment of KPIs with organizational vision, mission, and strategic goals scoring the highest mean. Overall problems and risks were at a moderate level, with data standards and management risks posting the highest mean score. Predictive analysis indicated that department affiliation was the most significant factor predicting the current state. Meanwhile, the key predictors for risk levels were KPI understanding, system access and usability, and systematic data storage, respectively. User segmentation identified three distinct groups:
1) High-Support-Required Group (37.13%), 2) High-Readiness Group (29.34%), and 3) High-Readiness with Risk-Control-Required Group (33.53%). In conclusion, organizations should transition from a standardized, single-policy approach to a phased rollout model, accompanied by targeted group-specific support aligned with the readiness and context of each unit.