76 Informatica Economică vol. 23, no. 2/2019 Software Effort Estimation Using Multilayer Perceptron and Long Short Term Memory Eduard-Florin PREDESCU, Alexandru ȘTEFAN, Alexis-Valentin ZAHARIA Bucharest University of Economic Studies, Romania [email protected], [email protected], [email protected] Software effort estimation is a hot topic for study in the last decades. The biggest challenge for project managers is to meet their goals within the given time limit. Machine learning software can take project management software to a whole new level. The objective of this paper is to show the applicability of using neural network algorithms in software effort estimation for pro- ject management. To prove the concept we are using two machine learning algorithms: Multi- layer Perceptron (MLP) and Long Short-Term Memory (LSTM). To train and test these ma- chine learning algorithms we are using the Desharnais dataset. The dataset consists of 77 sam- ple projects. From our results we have seen that Multilayer Perceptron algorithm has better performance than Long Short-Term Memory, by having a better determination coefficient for software effort estimation. Our success in implementing a machine learning that can estimate the software effort brings real benefits in the field of project management assisted by computer, further enhancing the ability of a manager to organize the tasks within the time limit of the project. Although, we need to take into consideration that we had a limited dataset that we could use so a real advancement would be to implement and test these algorithms using a real life company as a subject of testing. Keywords: software effort estimation, multilayer perceptron, long short-term memory, neural network algorithms, machine learning, Desharnais dataset Introduction within some different given constraints. Typi- 1 In this paper our aim is to prove the effec- cally, this include but are not limited to, esti- tiveness of machine learning in the planning mating a precise time frame in which the pro- and execution of a vast amount of projects. ject will be completed which can be one of the Machine learning, is a process of intense data trickiest parts of the job. Machine learning al- analysis that concludes with the automation of gorithms offer a simple solution to that conun- analytical model building. It is a branch de- drum as they can make project management rived from artificial intelligence which con- simpler and more efficient. Different machine sists of the fact that systems can learn from learning frameworks offer different benefits analyzed data, recognize patterns and make based on the particular field that they are be- calculated decisions with minimal or no hu- ing developed for. man interaction needed. Some of the advantages of machine learning The challenge that this paper undertakes is to are as follows: indicate which of the two machine learning al- Predict and assign tasks to team members gorithms studied MLP or LSTM is more effi- based on their skills; cient in the field of the machine learning in Predict and determine when deadlines project management. The direct result of the aren’t going to be met and what can be aforementioned algorithms is to simplify the done to prevent it; tasks of managers, to produce more precise Correct task time estimates as a means to predictions and finally and perhaps most im- update the project information and notify portantly to increase sales and efficiency the manager that budget and time frames within the target company. may require extension. The primary challenge that most project man- If implemented correctly and trained accord- agers face is the achieving of project goals ingly a machine learning software can reach a level of automation in which it automatically DOI: 10.12948/issn14531305/23.2.2019.07 Informatica Economică vol. 23, no. 2/2019 77 tracks and detects different problems and de- to be realized and correctly implemented var- lays caused by various unexpected factors. It ious objectives must always be considered and can also automatically detect delays within a treated accordingly, objectives such as: project and act according to what the situation Analyzing and identifying how workforce demands. This way the industry would see a effort and performance is estimated and total shift in the way projects are run and exe- carried out in large-scale distributed pro- cuted and this should lead to a positive impact jects; on overall team performance and efficiency. Analyzing the accuracy of the aforemen- Based on the evolutionary trajectory of ma- tioned effort and performance estimation chine learning algorithms in project manage- processes in large-scale distributed pro- ment they will fundamentally change the way jects; we as people think about the running and exe- Identifying and investigating the major cution of projects, the challenges that arise on factors that impact the accuracy of effort the way, in the form of technology and soft- and performance estimates in large-scale ware limitations is something that must be distributed projects. considered very thoroughly. The implications of undertaking the implementation of machine 2 Literature review learning algorithms in a large-scale legacy Machine Learning is an application of artifi- project can prove to be and insurmountable cial intelligence that is revolutionizing the obstacle if we are to consider the following way companies can do business. ML algo- problems: rithms allow programs to analyze big data to Hardware and software expenses, the amount increase the prediction power of the business of money required to be invested in both soft- which in return increase the efficiency of the ware development and hardware to train a business and its sales. software product can reach a staggering The biggest challenge for project managers is amount, due to the fact that machine learning to meet the goals of their projects within the algorithms are developed and trained with a given time limit. Normally, to predict the time particular objective in mind and quality of the frame of the project can be the hardest part of required hardware components. While not the job. Machine Learning can take project necessarily a barring problem for big corpora- management software to a whole new level. tions with huge profit numbers and ability to Many companies are currently testing Ma- invest in such a technology, it is still a time chine Learning to predict which actions to consuming operation which if not done cor- take. rectly can lead to serious problems down the There are already efforts of researching the ef- road. fectiveness of ML algorithms for project man- We aim to achieve this goal by implementing agement and our search has led us to a few and testing two machine-learning algorithms publications [2],[3],[4]. on the Desharnais dataset: Multi Layered Per- A case study analysed by Usman et al. on a ceptron (MLP) and Long Short Term Memory large-scale project concerned with the devel- (LSTM). In the following sections we will opment and maintenance of a telecommunica- cover the aforementioned machine learning tion software product at Ericsson shows that algorithms: although traditional estimation methods may The sections that will follow this part of our deliver reliable and accurate results, underes- study aim to provide and analyze real, tangible timation is a frequent trend at both quotation results. This will be realized though the con- and analysis stages of a software project esti- struction of a machine learning system that mation [5]. will take as input data from a variety of real Therefore, we can affirm that traditional esti- life projects, and will return the best possible mation techniques may be inefficient and im- result aimed at improving the planning and practical as outlined by Pospieszny et al. due execution of similar projects. In order for this to the fact that, often, pressure from clients or DOI: 10.12948/issn14531305/23.2.2019.07 78 Informatica Economică vol. 23, no. 2/2019 management or limited knowledge about the being a fundamental stage of a machine learn- domain or risks involved may lead to overop- ing method that has a large impact on the ac- timistic estimates, thus impacting severely curacy of the predictions as presented by project outcomes [3]. Huang et al. [1]. The same Pospieszny et al. suggests that with Furthermore, Panda et al. and Rijwani et al. smart data preparation and implementation of demonstrate that machine learning is a viable ML algorithms, we can obtain very accurate approach in software effort estimation, but it results compared to other approaches and are is highly dependent on the quality of the input suitable for deployment in practice, but for the dataset and the types of algorithms used for best results, it is recommended that homoge- the actual predictions [2][5]. neous data must be used for greater prediction Outlined below, table 1 presents the ap- accuracy [3]. proaches undertaken by the researchers in the Another context that needs to be addressed articles chosen as references for this paper to when implementing machine learning algo- validate machine learning as a viable software rithms is data, specifically data preprocessing effort estimation technique. Table 1. Short presentation of literature review Author Title Subject Argument Conclusion Pospieszny et An effective ap- Tackles the limi- Usage of the Traditional and al., 2018 proach for soft- tations and nar- ISBSG dataset parametric esti- ware project ef- rows the gap be- for modelling mation tech- fort and duration tween up to date with three ma- niques may be estimation with research in ma- chine learning inefficient and machine learn- chine learning algorithms for impractical. Pro- ing algorithms and potential de- cross-validation posed ML algo- ployment of ma- validates the rithms are suita- chine learning practicality of ble for estimat- algorithms in ML algorithms ing duration and practice for pro- in real world effort.