Principal Component Analysis on Morphological Variability of Critical Success Factors for Enterprise Resource Planning

The concept of critical success factors (CSFs) has been widely used as a measure to tackle the hurdles associated with numerous implementations of enterprise resource planning (ERP) systems. This study evaluates the morphological variability of CSFs using the analytical principal component analysis technique to identify principal components (PCs) that can be adopted for a successful ERP system implementation. The dataset of 205 CSFs from 127 different studies was evaluated for the morphological variability in those studies. According to the results, 66 PCs were identified and ranked accordingly. The first 49 PCs with eigenvalues greater than 1 accounted for 89.67 % of the variability recorded. The first 6 PCs respectively accounted for 13.67%, 19.37%, 24.67%, 29.41%, 33.52% and 36.94% cumulative variations. In general, the graphical illustration of the study results show the palpable division between the taxonomic groups for 3 PCs. Keywords—Enterprise resources; morphological variability; principal component; resource planning; success factor


I. INTRODUCTION
Critical success factors (CSFs) have been identified to be an essential precept for a successful ERP system implementation [1]. A critical success factor is a variable that has a significant impact on delivering a measurable improvement to project success [2]. The relevance of CSFs classification in ERP systems has been emphasised in various studies using different methods [1,3]. Certain authors, Hentschel, Leyh and Baumhauer [4] as well as Denolf, Trienekens, Wognum, van der Vorst and Omta [5], have postulated that despite the strong focus on avoiding failure in system implementation using CSFs approach, CSFs remain rarely researched. On the other hand, Saxena and McDonagh [6], have contended that CSFs remain the most-researched areas over the past years within the domain of enterprise systems.
However, despite this contention, there exists a consensus among researchers that CSF is a highly significant concept that can help address the inherent challenges associated with ERP system implementation [7]. Moreover, this has led to the identification of diverse CSFs in the literature. Consequently, the overarching objective of this study is to apply principal component analysis (PCA) to analyse morphological variability of CSFs for successful implementation of ERP systems. The realisation of the objective of this study affords the following distinctive contributions. An enhanced understanding of the concept of CSFs that acknowledges their morphological variability for an efficacious implementation of ERP systems. The application of a robust analytical method to provide valued acumen to the CSFs phenomena of ERP system implementation. The remainder of this paper is succinctly summarized as follows. The next section provides the background discussion with respect to the related literature. This is followed by the description of the material and methods of the study. Next is the presentation of results and discussion and the paper is briefly concluded.

II. BACKGROUND
The nature of CSFs has been reported in the literature to be inconsistent and repetitive, yielding the need for more analytical scientific methods [8][9][10]. Epizitone and Olugbara [11] highlighted this need by emphasising on the holistic nature of CSFs in different application settings. This view is further supported by the adoption of a mixed method research approach to tackle the complex phenomena of CSFs [12]. The determination of morphological variability of CSFs is a significant part of a successful implementation of an ERP project. The significance of CSFs classification has been emphasised in various related studies with a lot of attentions paid to the importance of CSFs and the success of ERP system implementations [9]. Consequently, the application of PCA to extract relevant information regarding CSFs from a large dimensional dataset is considered to enhance a deeper understanding of the intrinsic characteristics of CSFs [13].
PCA is a useful mathematical technique for emphasising variations and exposing hidden patterns in a dataset. It is predominantly applied for dimensionality reduction in application domains such as computer vision and pattern discovery in data mining [14]. It has been successfully used to specify principal components in varieties of datasets in many other areas of data science [14][15][16][17][18]. The technique has the potential to reveal essential characteristics while capturing the main structures of CSFs variability [19]. It is useful for discovering, reducing and identifying meaningful variables in a dataset. Hanci and Cebeci [15] have reported PCA to be a multivariate statistical technique with the capability of converting a lot of likely correlated factors into a set of smaller factors called principal components (PCs). The direction of the first PC is the same with the largest eigenvalue allied with its eigenvector. While the direction of the second PC is determined by the eigenvector, which is related to the second largest eigenvalue.
The PCA technique involves a mathematical procedure that is based on the eigen analysis, which computes eigenvalues and corresponding eigenvectors of a square symmetric matrix with sums of squares and cross products [20]. The paramount objective of this study as earlier stated is to apply the PCA technique to analyse morphological variability of CSFs for successful implementation of ERP systems. The analysis technique would help to identify different PCs for promoting ERP system adoption [21]. It is assumed that the results of this study will provide the knowledge of CSFs that is appropriate for use in a successful implementation of an ERP system.

III. MATERIAL AND METHODS
In this study, a total of 205 CSFs identified from 127 studies [22][23][24][25][26][27] was compiled and represented in a binary format displaying the feature of the identified variables for further analysis. The study dataset shown in Table S1 describes each factor as well as provides 205 qualitative CSFs and 127 quantitative instances that are suitable for PCA. The dataset was subjected to PCA to characterise the CSFs and identify the weight of each factor. The PCA technique was applied to a transformed dataset that was standardised into units of classes and attributes to determine the morphological variability. The number of PCs was determined using the minimal eigenvalue of unity called Kaiser criterion [28]. The dataset consisted of attributes 1 to 205 coded numerically as @ ATTRIBUTE F1-F205, while the related papers investigated for the extraction of factors were coded as @ATTRIBUTE class (P1-P127). All statistical procedures for the evaluation of morphological variability were obtained using the IBM SPSS statistics version 25 and WEKA 3.8.3. These statistical tools mutually afford an added validation advantage in identifying variations among the CSFs for ERP system implementation. The focus was on their morphological variation as it influences implementation success whilst providing the chance to analyse more than one factors in association. The Vendor (F1) extraction value for PC represents the lowest value of 0.631, while the maximum extracted values are for F31-Professional training services, F32-Setting realistic deadlines, F37-User participation in defying new processes, F58-Deep understanding strategy, F60-Former major change experience, F81-Business change is first to be considered, F85-Level of implementation acceleration, F139-Opportunities for growth and F146-Data model is compatible with data requirements. It can be noted from the first component that these factors loadings were integrated to account for the high eigenvalue.

IV. RESULTS AND DISCUSSION
The first 6 PCs cumulative variations are 13.67%, 19.37%, 24.67%, 29.41%, 33.52% and 36.94% respectively as shown in Table I. These PCs can be seen to be distinctively illustrated by screen plot in Fig. 2 Table II, present each factor loading used for extraction that can be seen within a range of 0.631 minimum to 0.995 maximum for the component extracted. Table II further shows the result of the analysis presented for the communality showing the contribution of each factor.
The PC one (PC1) has an eigenvalue of 28.015, which explains 13.667 as the total variance with the same value for the cumulative variance. Taking into composition the contributions of individual weighted factor values for the PC occurring from the factors in Table II. The contribution of 10 factors can be seen in the table identifying different groups for the 6 PCs (Table I). Fig. 5 shows the first six components in rotated space. The contribution can further be seen in Table S3. The first group for PC 1 includes CSF such as Business change is first to be considered with eigenvector of 0.166 variation that reflects environment to the level of implementation acceleration and using ERP to fulfil cross-functional areas with 0.15 variation. This component presents the largest variability in the dataset as compared to the subsequent components [15,29].       These results report the presence of great morphological variability for some of the CSFs presenting specification of the CSFs diversification of ERP system implementation based on the taxonomy of the groups possibly identified by the selection of these CSFs. In this paper, we have explained the morphological variability and tried to model the CSFs to diverse components that are relevant to ERP system implementation. It can be seen from these results that taxonomic groups were conceivably attained by selecting these features. Azadeh, Afshari-Mofrad and Khalojini [30] and García, Rivera and Iniesta [31] applied PCA to their studies to characterised CSFs. The current study explicates on the diversity of CSFs variability based on different identity groups. Many studies undertaken on CSFs have selected certain CSFs to contextualise their results. However, results of the current study are attained from the inclusion of all the identified CSFs to provide a holistic nature of CSFs with different morphology. A similar approach to Ahmad, Haleem and Syed [22], study where all CSFs identified were retained for further analysis [3], characterised CSFs using a hybrid approached of PCA and impact factor analysis to identify, validate, rank and classify factors as critical, active, inert and reactive. Bhatti [32] applied PCA on a smaller dataset consisting of data from 53 inputs, using the reliability and validity scale to explain and characterise 11 CSFs with eigenvalue greater than 1 that only assimilation factor loads greater than 0.5. Madapusi and Ortiz [33] report findings on ERP, discussing two factors that account for 50.315 of the variability following a lesser Cronbach alpha statistic of 0.60 as compared to Bhatt [32] who used 0.75.
The projection of the 205 CSFs morphology in the twodimensional graph of the component plot is shown in Fig. 3 and Fig. 4. The first, second and third PC coordinates of the PCA is realised using the morphological data accounted for 24.67% of the diversity observed (Fig. 3). While the subsequent three PCs four, five and six in Fig. 4 accounted for 12.30%. Overall, these displays denote an obvious division between taxonomic groups of CSFs relevant for the success of ERP system implementation.

V. CONCLUSION
Employing different markers of the CSFs, diversification was estimated by exploring the morphological attributes that provide essential preliminary method for gauging different CSFs while concurrently elucidating their performance under successful implementation. The substantial knowledge presented by the results of this study is the CSFs variability applicable to various implementations of ERP systems. In this study, 205 different CSFs were analysed by using data obtained from 127 studies presenting different morphological findings of CSFs. The low variability of the first six principal components demonstrates that the diversity of the pool was significantly with the highest CSFs having eigenvectors not limited to values such as 0.25, 0.205, 0.193, 0.169 and 0.157.
The results of this study provide an important contribution to the ERP CSFs body of knowledge with a special attention paid to the morphological features of a disparate model from several morphological taxonomies of the identified CSFs using a robust analytical method. The study results can help practitioners not to neglect any CSF, rather they should attach significant consideration to their roles in ensuring a successful implementations of ERP systems.

ACKNOWLEDGMENT
The first author would like to thank her fellow researchers for their inspirations and the Durban University of Technology for the support provided during the study.