An effective strategy for initializing the EM algorithm in finite mixture models

摘要: Finite mixture models represent one of the most popular tools for modeling heterogeneous data. The traditional approach parameter estimation is based on maximizing likelihood function. Direct optimization often troublesome due to complex structure. expectation---maximization algorithm proves be an effective remedy that alleviates this issue. solution obtained by procedure entirely driven choice starting values. This highlights importance initialization strategy. Despite efforts undertaken in area, there no uniform winner found and practitioners tend ignore issue, finding misleading or erroneous results. In paper, we propose a simple yet tool initializing setting. idea model averaging efficient detecting correct solutions even those cases when competitors perform poorly. utility proposed methodology shown through comprehensive simulation study applied well-known classification dataset with good