Keywords

1 Introduction

Fifth generation language or fifth GL even tends to be programming languages, which contains visual tools to develop a program. It utilizes visual and graphical advancement interface device to make the source language that is assembled with a third GL or fourth GL compiler. Visual programming enables you to see object-oriented structures and drag symbols to gather program squares. There are some important points as follows:

  • A fifth-generation programming language is a high level and logic language. User knowledge bases, expert systems, and less programmer control.

  • Fifth GL is a programming language dependent on tackling issues utilizing limitations given to the program, as opposed to utilizing a calculation composed by a software engineer.

  • Most imperatives-based, logic programming languages, and some decisive languages are fifth GL.

  • Fifth GL is utilized mostly in artificial intelligence inquire about.

Artificial intelligence (AI) is growing such a mind-blowing speed; occasionally, it appears to be supernatural. There is a supposition among analysts and designers that AI could develop so monstrously solid that it would be hard for people to control. People created AI frameworks by bringing into them each conceivable they could, for which the people themselves presently appear to be compromised. AI has many other areas as soft computing technologies, neural network, genetic algorithms, fuzzy logic modeling system, etc., for finding the precise expectation of software development cost estimation. The FIS technique has been adopted to predict maintenance cost.

2 Related Work

There is basic two category of models, such as algorithmic and non-algorithmic [1]. Everyone require inputs, a precise estimate of explicit traits, for example, lines of code and other cost drivers like range of abilities which are difficult to procure amid the beginning time of software development. In 1990s, non-algorithmic was conceived to extend estimating cost. Analysts have focus toward novel methodologies that delicate registering, for example, ANN, GA, and fuzzy logic [2, 3]. A portion of early works demonstrates that fuzzy logic offers an amazing etymological interpretation that ready to speak to imprecision in sources of inputs and outputs while giving huge learning ways to deal with model’s structure. It is a procedure to take care of issues, which are too complex to be in any way seen quantitatively. It depends on fuzzy-set-theory. It gives a system to speaking to semantic builds, for example, many, low, medium, and high. It gives a deduction structure that empowers fitting man reasoning limits. Unexpectedly, the binary set hypothesis depicts crisp events that do or do not happen. This experiments perspective hypothesis that clarify if the events will happen evaluating the opportunity for which the given events are required happening [4, 5].

It is the rule-based system which is the core learning containing the implied Fuzzy IF THEN guidelines in which a couple of words are portrayed by consistent part works. It can be classified into three kinds: pure, Takagi and Sugeno's fuzzy logic with fuzzifiers and defuzzifiers. The enormous bit for planning produces crisp-data as for input and envisions crisp-data as for output. This was right off the bat created by Mamdani. It has been viably associated with a collection of mechanical techniques and customer things [6, 7].

It is the initial phase in the fuzzy inference process. This includes an area change where crisp inputs sources are changed into fuzzy inputs. Crisp inputs are careful information sources estimated by sensors and go into the control system for handling, for example, temperature, weight, etc. [8,9,10]. Each crisp input that will be prepared by the FIS has its gathering of membership functions or sets to which they are changed. This gathering of membership functions exists inside a vast expanse of talk that holds every single important value that the crisp input can have. The accompanying demonstrates the structure of membership functions inside a vast expanse of talk for a crisp input [11, 12].

The principal of FIS is concentrated at the capacity of fuzzy logic that show characteristic. This system contains fuzzy-rules worked for expert-knowledge and called fuzzy expert systems, contingent upon their last use. Before FIS, it was at that point applied to construct expert systems for recreation objectives [13, 14]. The master frameworks depended upon the classical-boolean-logic that was not appropriate for dealing with the sequentially to the fundamental procedure wonders. Fuzzy-logic enables continuous standards that be brought into expert-knowledge-based test systems [7, 15].

The Sugeno's initial tasks, a great deal of scientists, have been engaged with structuring fuzzy systems from databases. The means of fuzzy reasoning performed by FISs are as follows:

  1. 1.

    Comparison of input factors with the MF on the precursor portion to get the membership estimations of each phonetic mark. (Progression is frequently said fuzzification).

  2. 2.

    Connection of the membership values on the reason portion to get terminating quality (level of satisfaction) of each standard.

  3. 3.

    Production of certified ultimately (either fuzzy or crisp) or each standard relying upon terminating quality.

  4. 4.

    Composite the certified consequents to deliver a crisp output. (Progression is said defuzzification).

3 Research Methodology Used

The FIS is used to execute the differing preparing propels. Decisions were obliged making and adjusting FIS with fuzzy logic toolbox software using graphical instruments or command line capacities. The research will implement on third GL, fourth GL, and fifth GL using Mamdani FIS. Figure 1 is used as fuzzification and defuzzification.

Fig. 1
A framework. Crisp input is given to fuzzification unit, moves to decision making unit and de-fuzzification unit, and receives a crisp output. A knowledge base consisting of database and rule base shares data to both fuzzification and de-fuzzification units.

Fuzzy inference system (FIS)

Full size image

The performance analysis and their corresponding results are compared. The results are analyzed using the criterion RMSE. Less value means that the result is more accurate.

  • Select a specific kind of FIS (Mamdani).

  • Define the variables for the input and output.

  • Set input and output member functions.

  • The data is now in rule editorial manager if-then rules.

  • An explicit model structure is made, and parameters of input and output factors tuned to get the ideal output.

4 Proposed Model for Software Maintenance Cost Estimation

COCOMO II is utilized as the model-based to assess the software project cost. The model was developed by Mr. Boehm and Scattered in 1981 that utilizing aggregated data from 63 projects. It is a good manual that estimating maintenance cost for software. This proposed model implies with new factual methodologies and strategies that estimate the maintenance cost of software using fifth GL (fuzzy inference system) procedure.

The issue of software cost estimation is that everything considered and relies on single estimations of size, cost drivers, and scale factors. It is assessed dependent on recently finished projects that are fairly like the present projects [16,17,18]. Similarly, cost drivers and scale elements need through evaluation instead of doling out a fixed numbers value. To beat this condition, it is more brilliant to address these responsibilities to the kinds of fuzzy-sets, where the qualities of interval are utilized which is expressed through collection of membership functions like triangular MF, trapezoidal MF, and Gaussian MF [19].

The proposed fuzzy software cost estimation model is represented in the Fig. 2. Its principles contain phonetic factors identified with the undertaking. The FIS utilizes connecters “and/or” for COCOMO input factors that shape principles. The FIS incorporates many input software characteristics: seventeen cost drivers, five scale factors, one size (KDLOC), and one output as cost estimation (CE).

Fig. 2
A flowchart of 5 blocks. 1. Software maintenance phase. 2. Size, cost drivers, scale factors share data for fuzzification. 3. Cycle of model generation consists of F I S techniques, optimization of model, membership function. 4. Performance analysis based on R M S E factor. 5. Estimated cost.

Proposed maintenance cost estimation model using FIS techniques

Full size image

Fuzzy set takes all the input and convert into the phonetic values. For each cost driver, a different FIS is planned. Principles are created as cost drivers for forerunner parts that comparing effort-multiplier in the resulting portion. The defuzzified value for the effort-multiplier has kept for separate FIS. The scale factors are additionally fuzzified. The pcap means that programmer-capability is examined for an example. Programmer-capability for fuzzification depends on COCOMO II, calibrated model of post-architecture qualities.

Next, the model so obtained will be later subjected to optimization of its model parameters using fuzzy inference system optimization technique to arrive at better software cost estimation prediction accuracy. The fuzzy operators such union, intersection, and complement shall be used. FIS races to create answers for progressive generations. Henceforth, the quality of the solutions in progressive generations improves. The procedure is ended when an ideal solution is found. The result has analyzed the criterion of root mean square error (RMSE) factor, which predicts the better software cost estimation with accuracy.

4.1 Data Used for Validation

The data used as input and output variables for ideal COCOMO II model advancement is given in Table 1. The dataset Table 2 is assembled from the examination of 40 software projects, which is adopted from Software Engineering Repository of PROMISE dataset which open access for researching reason. It comprises 26 attributes like seventeen standards COCOMO II characteristics cost drivers and five scale factors in the range that measure in thousand delivered source lines of code (KDLOC) directions. The output of the model is the cost estimation (CE), measured in man-months. The estimated efforts using third GL, fourth GL, and fifth GL approaches obtained are tabulated and compared. The model equation is given as follows:

$$ {\text{PM}} = A \times \, \left[ {{\text{Size}}} \right]^{{{1}.0{1}}} + \mathop \sum \limits_{i = 1}^{5} {\text{SF}}i \times \mathop \prod \limits_{i = 1}^{17} {\text{EM}}i $$
(1)
Table 1 COCOMO II effort multipliers [5]
Full size table
Table 2 Estimated-effort using that different MFs
Full size table

Here, effort is indicated in terms of person-months (PM), A is a constant that is multiplicative, size is the projected-size of the software that expressed in KDLOC, EMi (i = 1, 2, 3, 0.17) are effort-multipliers, and SFi (i = 1, 2, 3, 0.5) are scale factors as exponent. It is a specific normal for product improvement that has impact that increments or decrements the measure for advancement effort [20]. There are team cohesion, process maturity, architecture/risk resolution, development, flexibility, and precedent Ness. All the effort-multipliers are gathered into 4 parts, which are project-factors, personnel, platform, and product. The items are utilized to modify the nominal effort.

4.2 Fuzzy Inference System Rules Applied

Fuzzy principles for the fuzzy inference system dependent on COCOMO II are characterized with semantic factors in the fuzzification procedure. These principles depend upon connective “and” as between the input factors.

The Rules are defined as follows:

if (rely is vl) then (effort is vl)

if (rely is l) then (effort is l)

if (prec is vl) then (effort is xh)

if (pmat is vh) then (effort is l)

The following rules are used in Figs. 3, 4, and 5:

Fig. 3
A graph of lines forms a Gaussian M Fs such as H, N, L, and V L. A line V H follows a decreasing flow intersecting H and falls flat.

PCAP fuzzification cost drivers using Gaussian MF

Full size image
Fig. 4
A graph of lines forms a trapezoidal M Fs such as H, N, L, and V L. The V H line drops steeply intersecting H, and remains constant.

PCAP fuzzification cost drivers using trapezoidal MF

Full size image
Fig. 5
A graph of lines forms four triangular M Fs. A line from the left labeled V H sharply declines to the start of line N, and becomes stable.

PCAP fuzzification cost drivers using triangular MF

Full size image

if (pcap is very low) then (increased effort)

if (pcap is low) then (increased effort)

if (pcap is nominal) then (unchanged)

if (pcap is high) then (decreased effort)

if (pcap is very high) then (decreased effort)

Figure 6 expresses the graphical user interface that developed our model FIS. We can legitimately enter the qualities and get the relating effort. The studies have been carried out using MATLAB simulation environment.

Fig. 6
A screenshot is titled fuzzy cocomo 2. The required data includes enter size, scale factors, cost drivers, calculation of efforts.

Interface used for cost evaluation

Full size image

5 Experimental Results

5.1 Root Mean Squared Error (RMSE)

The assessment comprises in contrasting the exactness of the calculated cost with genuine cost. There are numerous assessment scales for estimating software cost. We connected the regular one is RMSE and defined as follows:

$$ {\text{RMSE}} = \left( {\frac{1}{N}\sum\limits_{i = 1}^{N} {\left( {y_{i} - \mathop y\limits^{ \wedge }_{i} } \right)^{2} } } \right)^{{}} $$
(2)

5.2 Mean Absolute Error (MAE)

It is a proportion of expectation precision of a forecasting method in insights, for instance, in pattern estimation. The most part communicates precision as a rate and is characterized by the equation:

$$ M = \frac{1}{n}\mathop \sum \limits_{t = 1}^{n} \left| {\frac{{A_{t} - F_{t} }}{{A_{t} }}} \right| $$
(3)

The software effort got when utilizing regular COCOMO II and fuzzy MF were looked at. In the wake of breaking down, the outcomes acquired utilizing applying third, fourth, and fifth GL. It is shown that the cost evaluated by fuzzifying all effort-multipliers utilizing fifth GL (FIS) procedure is predicting better estimate.

6 Comparison

The parameter of cost estimation models for the assessment is the MAE that is represented in the Eq. 3. The effort has been calculated for every observation (Table 3).

Table 3 Comparison of MAE values
Full size table

Table 4 demonstrates with the chart representing to the similar examination of the real cost with the estimated cost using COCOMO II, third, fourth, and fifth GL. The RMSE and MAE values are calculated using Eqs. 2 and 3, and the RMSE values for all project for COCOMO II, third GL, fourth GL, and fifth GL are 1.2403, 1.0638, 1.075, and 0.9398, respectively. The MAE values are 0.1650, 0.1251, 0.1236, and 0.1183, respectively. This plainly demonstrates here is a reduction in the absolute errors; therefore, the proposed model is progressively reasonable for estimating cost.

Table 4 Comparison of RMSE and MAE factors
Full size table

7 Conclusion

We conclude that the use of fuzzy logic in SCE yields more exact outcomes than the past experimental model methodology. The RMSE values of cost estimation using fifth GL based on FIS techniques give better outcomes for most extreme rules if other high-level language techniques will be used. It found that the FIS is achieving better as it shows a likely change in its intervals, and accomplished outcomes were nearer to the actual cost. FIS has the lowest MAE and highest accuracy of the three generation language software methodologies that studied.

Future work incorporates more current procedures, i.e., type-2 fuzzy-system can likewise be connected for increasingly precise forecasts of software. The above research work can be easily employed in the software industries.