Evaluation of Asset Performance Using Integral Data Mining Scheme

The asset evaluation is an indispensable process in asset management and asset disposal. Seldom studies were found in exploring asset-usage performance of laboratory equipment in universities. The influential parameters, such as source, usage, and benefit that influence the performance of asset management in universities should be considered more structurally and deeply. In this study, a structural and complete index system and an integral data mining method are proposed to investigate the effect of related index parameters on the asset-usage performance of laboratory equipment in universities. The proposed integral method includes two phases. The first phase is to find out the relative weight of each influential parameter using the integral Analytic Hierarchy Process (AHP) scheme which extensively contains the index-weight and time-weight models based on Orness measure in conjunction with the method of relative closeness degree. The second phase is to investigate the global correlation between second-level indexes and the final performance using Grey System Theory (GST). The results of case study indicate that the proposed integral AHP-and-GST method can not only be used to globally and locally understand the effect of influential parameters on fixed asset of laboratory equipment, but also provide a decision-making basis for asset allocation.


I. INTRODUCTION AND LITERATURE REVIEW
Usually, the asset evaluation is used for pricing and asset transaction. The asset evaluation refers to assessment and estimation of its value before asset transaction transfer or change its operator. The asset evaluation is an indispensable process in asset management and asset disposal [1]- [4]. With highly attentions on the development of higher educations all over the world, governments invest much more capital in universities or colleges than ever. The major spend of college's capital lies in various assets. For a college featured with science and engineering departments, the asset of laboratory equipment takes up a major part of total funds. Conventional ways of arbitrarily allocating the asset of laboratory equipment often cause problems of asset idle, preemptive occupation, and so on. And, these matters result in the reduction of management performance of school The associate editor coordinating the review of this manuscript and approving it for publication was Senthil Kumar . asset. Under the constraint of limited finance budget for universities, how to efficiently use the laboratory equipment to cultivate students and perform science studies are essential. Finding a proper way to efficiently allocate the asset of laboratory equipment is an important issue in the operation and management of a university. In the past, a large amount of studies concerned problems of asset allocation [5], asset management [6], asset risk [7], and etc. in many application fields such as engineering, company, and individuals. Seldom papers were found in dealing with the asset performance evaluation of school's laboratory equipment. Thus, it is expected to find a good asset performance evaluation method for the usage of laboratory equipment in universities.
Among past proposed asset evaluation methods, there were three main streams: cost method, revenue method, and market method [8]. These methods all concerned about using the macro level scheme to evaluate assets. Some of them proposed simple or linear mathematical formulas to express the relationship between asset performance and asset usage. Besides, many studies used the weight or modified weight methods to deal with the asset evaluation problems [9], [10]. These methods were ambiguous and insufficient to describe the complex non-linear relationship between asset-related parameters and the asset-usage performance. Besides, assetrelated parameters, such as source, usage, and benefit that significantly influence the asset performance of laboratory equipment in universities, should be considered more structurally and deeply. Therefore, this study attempts to propose a structural and complete asset index system and an integral data mining method to investigate the effect of asset-related index parameters on the asset-usage performance of laboratory equipment in universities. In detail, a novel integral data mining method which includes the integral Analytic Hierarchy Process (AHP) scheme and the Grey System Theory (GST) is proposed to analyze and establish the relationship between the fixed asset performance and its influential indexes for laboratory equipment in universities.

II. PROBLEM DESCRIPTION AND FORMULATION A. CONSTRUCTION OF INDEX SYSTEM
The asset investment of laboratory equipment (LE) is crucial for cultivation of university students. In view of the existing problem of asset allocation of LE in universities and the main factors that affect their operational performance, this study attempts to propose an effective methodology to construct a performance evaluation index system of LE, which majorly includes four-aspect indexes: LE cost, LE fund rate, LE usage rate, and LE benefit. Further, based on the necessity for cultivating students' application abilities after using LE, we consulted twenty experts who worked at asset departments of different universities and all had plentiful experiences in asset managements. And, we obtain twelve second-level performance evaluation indexes which are explained as follows and listed in Table 1.

1) LE COST
The LE cost for a university is divided into three categories according to the amount of investment (called C 11 , C 12 , and C 13 ) in which, C 11 denotes the low-value equipment that has an amount lower than CNY 10000, C 12 denotes the middle-value equipment that has an amount between CNY 10,000 and 100,000, and C 13 denotes the middle-value equipment that has an amount over than CNY 100,000.

2) LE FUND RATE
The LE fund rate means the ratio of LE fund from certain source to total funds of school in the same year, and it can be divided into three categories as follows. C 21 denotes the fund rate from school. C 22 denotes the fund rate from government. C 23 denotes the fund rate from private enterprise.

3) LE USAGE RATE
The LE usage rate includes three terms as follows. C 31 denotes LE quantity over student numbers in a whole year. C 32 denotes the ratio of students' practice hours in using LE to students' total learning hours in a whole year. C 33 denotes the undamaged rate which is the ratio of undamaged numbers to the total numbers of LE in a whole year.

4) LE BENEFIT
The LE benefit includes three categories as follows. C 41 denotes the project numbers of using LE in a whole year. C 42 denotes the technological prize-winning numbers of students who ever used LE in a whole year. C 43 denotes the grant authorization numbers of patents from school to enterprise in a whole year.

B. PERFORMANCE EVALUATION MODEL OF LABORATORY EQUIPMENT ASSET
In the first phase, we use the integral AHP scheme which contains the conventional AHP but integrates with the relative closeness degree as the judgement evaluation index to investigate the influence of various indexes of LE asset on the asset performance. In the second phase, we use the GM(1,N) model of GST scheme to obtain a non-linear relationship between indexes of LE asset and asset performance. VOLUME 8, 2020

1) AHP METHOD
To deal with the asset performance evaluation problem, several different methods have been proposed such as neural network [11], [12], analytic hierarchy process [13], fuzzy logic [14], and etc. The most extensively used method is AHP decision-making method by Satty in 1980. AHP is a powerful tool in solving decision-making problems and has been successfully applied to many fields [15]- [17]. It enables decision makers to handle unstructured and complex problems in a hierarchy way for determining the priorities in a systematic manner. AHP uses pair-wise comparisons and derives priorities among all the criteria and sub-criteria within each level. AHP combines the criteria weights and option scores to obtain a global score for a consequent ranking. The higher weight means more important the corresponding criterion is. And the higher score means the better performance of the option is.

2) GREY SYSTEM THEOREM (GST)
GST, firstly proposed by Julong Deng in 1982 [18], is a method that focuses on the investigation of problems involving small samples or poor information. It may be used to deal with uncertain systems that have partially known data. Through generating, excavating and extracting useful data, the behavior of system can be correctly described and effectively monitored. As known, there always exist lots of uncertain systems with small samples and poor information. This fact motivates the wide range of applicability of GST.
In this study, the GM(1, N) scheme of GST [19] is used to establish the correlation relationship between LE asset indexes and final asset performance in terms of the influence weighting. Basically, the GM(1,N) model adopts a first-order differential equation to establish the relationship between independent and dependent variables. Followings are a brief introduction of GM(1,N) algorithm.
3) Establishing the correlation relationship: It is assumed that a first-order differential equation holds true for the above two sequences: Equation (4) where y (1) Equation (5) can be rearranged into a matrix form: where

5) Calculating Q
The matrix Q can be obtained by using the least-square error method as Hence, the influence weighting of the major sequence (independent variables) on the target sequence (dependent variables) can be obtained by comparing the norm values of q 2 ∼ q N .

3) CALCULATING INDEXES OF RELATIVE CLOSENESS DEGREE
Due to insufficient ability for using traditional weighting method to handle evaluation problems, we now introduce the Index of Relative Closeness Degree (IRCD) method to more reasonably and accurately evaluate the performance of fixed asset of laboratory equipment. Conceptually, IRCD means how close or how much influences of the index parameters on the final target parameter. IRCD is comprised of normalized weights in each layer and defined as follows.
The IRCD in t th year of the first-level index C i for certain school a s is denoted by D sti and calculated as in which D + sti and D − sti are respectively defined as follows: Similarly, the IRCD in t th year of the second-level index C i for certain school a s is calculated as The final IRCD for certain school is denoted by D s and calculated by the following formula: Figure 1 shows the architecture of our proposed methodology. First, we investigate and formulate the problem under consideration. Second, we construct a two-layer index system that includes four first-level indexes and twelve second-level indexes based on the survey from experts. Third, we perform questionnaire surveys to acquire the necessary data of assessment. Fourth, we use the integral AHP and GM(1,N) models to respectively analyze and establish the correlation of the asset performance of LE. Fifth, we make analyses, comparison, and discussion. The calculation details of integral AHP are introduced as follows.

III. MANIPULATION PROCEDURE
Step 1: Calculation of consistency ratio According to the principle of AHP mentioned before, we first proceed the evaluation of C ij by experts using Delphi method with 1-9 points of judgement scale (shown in Table 2). The linguistic evaluation results of experts' judgements for each criterion are transformed into pairwise comparisons. Then we may construct a pairwise comparison matrix A in which the intensities of importance from activity i to j is denoted by a ji (i < j =1,2,3,. . . ,n), and components a ji = 1/a ij denote their reciprocal numbers.
The geometric averages of weights of second-level indexes C ij are defined byw ij = n n k=1 a ijk .
The normalized weights are defined by Hence, the weights of first-level indexes are expressed as w i = w i1 w i2 ... w in T . The consistency index (η ci ) is defined as [16]: where n is the order of judgement matrix A and λ max is its maximal eigenvalue. The consistency ratio (η cr ) is defined as η cr = η ci /η ri , where η ri is the random average consistency VOLUME 8, 2020 index, as shown in Table 3. The decision can be deemed if η cr < 0.1 and the judgement is inconsistent if η cr ≥ 0.1.
Step 2: Normalization of evaluation results To eliminate the influence of different magnitudes and units on decision results, a normalization formula is used to transform x st,ij to x st,ij , defined as follows: Step 3: Calculation of IRCD The IRCD in t th year of the first-level index C i for certain school a s is calculated via equations (10)- (12).
Step 4: Calculation of time weight using maximal entropy method.
To calculate the time weight in t th year of the related indexes, we establish an optimal problem as following, in which it may be solved based on the maximum entropy method.
where the magnitude of γ means the importance of time sequence in viewpoints of decision makers. The usual values of γ are {0.1, 0.2, 0.3, 0.4, 0.5} which respectively represent extremely important, strongly important, obviously important, slightly important, and equally important in viewpoints of decision makers for all data in all periods of time. In this study, we choose the middle value of γ = 0.3.
Step 5: Calculation of final IRCD The final IRCD for certain school a s can be calculated by equations (13)- (16).

IV. CASE STUDY AND DISCUSSION
This section includes the school background, problems in school, and implementation of integral AHP.

A. BACKGROUND AND PROBLEMS IN SCHOOL
Our investigation data were obtained from four different universities, denoted by a s , b s , c s , and d s , in Fujian Province, China. For the sake of similar comparison bases for all cases, we only choose the engineering colleges from these schools, which are: the college of mechanical and electrical engineering in a s school, the college of mechanical and car engineering in b s school, the college of mechanical and automation engineering in c s school, and college of mechanical and intelligence manufacturing in d s school. These colleges had  Table 4. These data are normalized via equation (21) and the calculation results are shown in Table 5.
Second, according to the survey by experts in conjunction with the results from calculation via equations (17)- (19), we obtain weights of the first-level W 1i and second-level indexes W 2i , as shown in Table 6. The consistency index is calculated via equation (20) and obtained as η ci = 0.059. It is judged that the calculation result of weights satisfies the consistency requirement.
Third, based on previously obtained weights (listed in Table 6), we obtain the first-level indexes D sti for four different colleges through calculation via equations (10) and (11). And the results are shown in Table 7. Then, we may construct an optimal evaluation weight problem as follows: The above optimization problem is solved by using LINGO software, and the final synthetic weights are obtained as:    Apparently, it is deduced that in which it means that the synthetic performance sequence for fixed asset of LE among schools a s , b s , c s , and d s in years of 2014-2017. As can be seen from equation (25) that universities a s and c s have better usage performances than universities b s and d s for the asset allocation of laboratory equipment. Meanwhile, it is seen from equation (26) that university a s has the best performance for the asset allocation of laboratory equipment, while university d s has the worst.
From the above in-depth investigation, we summarize that: 1) The reason of high performance score for university a s is that it has obvious achievements in student cultivation than others. However, the final synthetic score of university a s is only slightly greater than university c s because of less financial investment in the laboratory equipment.
2) The university c s has the highest financial investment in the laboratory equipment and meanwhile it has a better average performance index than other universities. However, owing to the apparently weak achievement of student cultivation, university c s behaves a little bit worse than university a s eventually.
3) Both universities b s and d s are insufficient for the financial investment of laboratory equipment. This results in worse asset-usage performances for universities b s and d s than for universities a s and c s . However, due to better behaviors in the equipment usage and benefit, the synthetic score of university b s is higher than the university d s . 4) From equation (16), we may obtain the differences between final IRCD (d s ) of universities a s , b s , c s and d s as 0.0003, 0.1142, 0.1495, respectively. It is seen that if the financial investment in laboratory equipment for university a s increases, it will cause a great increase in synthetic score than that for university c s . On the contrary, if the achievement of student cultivation for university c s increases, it will cause a slight increase in synthetic score than that for university a s .

C. CORRELATION ESTABLISHMENT USING GST
To quantitatively study the synthetic performance for the fixed asset of laboratory equipment among schools, we now VOLUME 8, 2020 introduce the GM(1,N) scheme whose details were mentioned before. Considering the data of second-level index values listed in Table 4, we first take averages of C i in years 2014 to 2017, denoted asC i . ThenC i is normalized toĈ i on the basis of their maximum values. A questionnaire survey, using second-level indexesĈ 1 ∼Ĉ 12 as independent variables and the performance of fixed asset of laboratory equipment (AP) as dependent variable, was done to thirty related professional employees of universities. The results of target AP was estimated in the form of semantic-differential method using credits with 1∼7 points. Further normalization of AP is carried out and it is found that the sequence of AP keeps the same as that in equation (26). However, this time we obtain a subjective and meaningful credits of AP.
In constructing the correlation between second-level indexes and AP using the GM(1,N) scheme, we set z 1 as AP and z 2 ∼ z 13 asĈ 1 ∼Ĉ 12 in equation (4) and solve equations (5)∼ (9). A correlation result based on the a firstorder ordinary-differential-equation is obtained and shown in Figure 2, where coefficients q 2 ∼ q 13 mean the influential weightings ofĈ 1 ∼Ĉ 12 on AP. Considering the top three, it is seen thatĈ 9 , defined as the undamaged LE, has the largest influential weighting (q 10 ) on AP. AndĈ 6 , defined as the fund rate of private enterprise, has the second influential weighting (q 7 ) on AP. AndĈ 4 , defined as funded by school, has third influential weighting (q 5 ) on AP. If we increasê C 9 ,Ĉ 6 , andĈ 4 , it will cause an apparent increase in AP. Furthermore, the last three indexes:Ĉ 12 ,Ĉ 11 , andĈ 10 that have weightings q 13 , q 12 , and q 11 respectively, have minor effects on AP comparing to other indexes. Investments on the last three terms are less effective. The influential sequences of all second-level indexes on AP can be expressed as q 10 > q 7 > q 5 > q 8 > q 6 > q 9 > q 3 > q 2 > q 4 > q 13 > q 12 > q 11 (27)

V. CONCLUSION
This study proposes a complete and structural asset index system as well as an integral data mining method which include the integral AHP and GST schemes to investigate the performance and find out the correlation for asset usage of laboratory equipment in universities. We use the hierarchy process (AHP) scheme in conjunction with the method of relative closeness degree to globally evaluate the asset performance of different universities. And, the GM(1,N) of the grey system theory is used to locally investigate the correlation between second-level indexes and the final performance.
The results of case study show that the university a s has the best AP among all schools and the first three influential second-level indexes on AP are:Ĉ 9 (undamaged LE),Ĉ 6 (asset funded by private enterprise), andĈ 4 (asset funded by school). The proposed methodology is an innovative and effective scheme to analyze the concerning issue of asset performance of laboratory equipment in universities both globally and locally.