近年來國內大型工程逐漸飽和，走向國際儼然成為工程技術顧問業重要的發展方向。隨海外案量之增加，海外派遣人力需求亦隨之大增，然各公司卻面臨多數員工外派意願低落之況狀。過往研究指出，合適之外派人選將是海外專案成功關鍵，其中員工外派意願影響甚鉅，故本研究欲建立員工海外派遣意願之預測規則，協助企業了解員工考量並快速篩選合適外派人選。本文透過文獻回顧與專家訪談初步確立22項影響因素，並以具代表性之大型工程技術顧問公司之員工作為問卷發放對象，共發放1000份問卷並回收486份問卷，其中413份為有效問卷。 本研究利用約略集理論分析影響員工外派意願之重要因素，萃取出外派孤寂感、孩童照顧責任、具吸引力之回任方案等六項核心影響因子以及九項中等影響因子，而後本研究將出現率大於50%的10項因子視為具有顯著影響變數，並納入5項人口屬性作為員工外派意願之預測因子，共計15項分類變數，並以員工之外派意願做為決策變數，透過決策樹演算法建構工程顧問業員工外派意願之預測規則。分析結果顯示，拒絕外派規則共7條以及接受外派規則5條。透過預測規則可快速篩選有意願的外派人選後再進行面談，不僅節省時間也能降低搜尋成本。此外，本研究進一步發現，員工重視孤寂感程度以及員工婚姻狀態可做為企業判斷員工外派意願之快速指標。 ;The objective of this research is to identify and classify the factors affecting the expatriation willingness (EW) of engineering consulting company employees. Thirteen EW impact factors are summarized from the literature review and divided into four categories. From the collected factors and expert interviews, 22 impact factors are obtained and divided into eight categories, with the exception of demographic variables. A survey aiming at the top five engineering consulting companies is carried out. Out of a total of 1,000 questionnaires sent out, 41.3% valid responses are returned. The statistical analysis shows that the survey is reliable and one of the 22 factors is removed. The rough set theory (RST) is utilized to classify these factors into three classes based on the impact level. The conclusions provide practitioners with six core impact factors, nine medium impact factors and six Insignificant Impact factors on employees’ EW. Among them 15 factors are set as the inputs to establish prediction rules.
This paper describes the use of the recently developed SOM-based Optimization (SOMO) algorithm to determine the optimal parameter settings for a neurofuzzy classifier for dealing with a practical expatriation willingness (EW) problem. The results show that the SOMO neurofuzzy classifier yields 6 determination rules, one for positive EW and the rest for negative EW. Loneliness and marital status are the most significant attributes for deciding on personal EW for international projects. They both have high coverage and accuracy rates greater than 80%. Compared with C5.0 algorithm, we conclude that the proposed model apparently outpaces the C5.0 algorithm in terms of accuracy and coverage. SOMO is effective and efficient for optimizing parameter selection.
