Abstract. The high risk involved in the Scrum-based software development project comes from the variety of uncertainties that exist in each of its components. Therefore, the success rate needs to be predicted as a basis for the Scrum team to formulate an appropriate management strategy. This stochastic problem is represented formally in the (non-parametric) Bayesian networks model. We then design several scenarios to the generated large-scale Scrum-based development projects with multiple stakeholders and multiple feature teams. We tried to simulate several variables used in this model by using (rank nodes)-based as well as (survey and weight functions)-based algorithms. The experimental results show that the proposed model is running well so that it can be an alternative for Scrum team in predicting the success rate of the Scrum-based software development project.

Abstract. On a tour activity, travel time estimation is needed so that the travel itinerary goes according to the plan. Travel time estimation is very important so we can estimate the time needed to arrive at the destinations in the travel itinerary. Therefore we need a method that can estimate travel time from one place to another. In this study, we propose the k-Nearest Neighbors Regression (kNN-Regression) method with Tensorflow to construct an estimation model. The proposed number of features in our estimation model is 8 features, i.e. zone information, time information, day information, weather information, temperature information, wind speed information, humidity information, and precipitation information. The data obtained from travel information from Ngurah Rai airport to Kuta Beach using GPS and weather information using weather application in real-time. We divide our data into two groups: a historical group consisting of 177 data and a testing group consisting of 51 data. In the testing stage, kNNRegression will find the historical data closest to the testing data, so that the estimation value of the travel time of some testing data is not much different from the value of the nearest historical data. As a result, our proposed model gives the Mean Absolute Error (MAE) of 2.196078, Root Mean Square Error (RMSE) of 2.977036294 and accuracy rate 88.1819%.

Abstract. The occupants of a multistorey building are very vulnerable when disaster happens (e.g. fire, earthquake, etc.). In the worst case, the only way to escape is through the emergency stairways, in which they scramble with the occupants coming from the higher floors. This situation often creates bottlenecks along the way, which in turn extends the evacuation time. In addition, the panic situation increases the risks of accidents, causing injuries or even death. In this research, we use NetLogo to simulate the evacuation process from a simplified multistorey building based on some parameters, which are the placement of the emergency stairways, the number of occupants in each floor, the number of stories, and the height of the floor. We set simulation scenarios to analyze the relation between the number of stories with evacuation time in each configuration. In our experiment context, we conclude that when the time-limit applies, the total evacuation through stairways is not effective for buildings having more than three storeys.

Abstract. Performance of credit scoring model is a main concern for financial institutions in determining the credit risk of credit applicants. Credit score will be one of basis for the lender to make a decision, approved or rejected, for any credit applications. There are many methods and approaches that have been modeled for this problem. This study tries to explore further the Hill-Climbing Bagged Ensemble Selection (HCES-Bag) algorithm which has the best performance for credit scoring model as has been analyzed comprehensively in the research conducted by Lessmann et al. 1. We modify some average formulas for the base-level models to find out the opportunity for improving the performance of credit scoring model as measured by several performance indicators. Experiment with German Credit Data from the UCI Machine Learning Repository by first using Multivariate Adaptive Regression Splines (MARS) model for features selection demonstrates that the modification average does not affect credit scoring model performance significantly. However, some of them make the credit scoring model become more efficient because we can obtained same level of credit scoring model performances by using only smaller number of base-level models.

Abstract. This study aims to determine the model for optimal stocks allocation and normal contribution that can minimize the funding variation based on stocks returns and dynamic mortality rates in a Defined Benefit Pension Plan. In this study, assets are allocated to the stock market, as investments in the stock market can increase funding variation that lead to high risk of decreasing funds as well as lack of funds in paying Pension Benefits to participants. The optimization model used in this study is a model which the objective function is a quadratic function. The stocks used in this study were SMRA, PWON, GGRM, INTP, UNTR, UNVR, BBTN, PTBA, SCMA and ANTM, in addition to mortality rates using probability death data of female civil servants with age ranges from age 52 to 60 in 2008 to 2015. By using the optimization model, the proportion of SMRA is 6.59%, PWON is 19.42%, GGRM is 3.54%, INTP is 7.32%, UNTR is 8.03%, UNVR is 18.87%, BBTN is 16.71%, PTBA is 6.51%, SCMA is 9.07%, and ANTM is 3.94% and also normal contribution is Rp 20,976,310.

Abstract. There is a fundamental difference between image and gesture recognition where image recognition only works against one frame while gesture recognition works on a sequence of frames. It means that the accuracy formulas implemented on each issue are different. The accuracy of the image recognition is calculated based on the prediction accuracy of each frame, while gesture recognition is based on each sequence of frames. The incompatibility of using these accuracy formulas generate the misleading outputs and interpretation. Thus, the classification model used also needs to be adjusted with this problem. In this paper, we use GLVQ model as a classification algorithm based on machine learning approach to recognize the gestures of Indonesian sign language (BISINDO). However, this algorithm is used to classify every single frame so it needs to be modified by adding a new function for a sequence of frames, e.g. mode. In addition, there is a parameter known as the number of prototypes that affects the accuracy of the model. Based on the results of this research, GLVQ model with mode function has a higher degree of accuracy when compared with Hidden-Markov Model (HMM) in recognizing BISINDO. However, it is necessary to specify a more appropriate function instead of mode which is not give uniquely results. We also know that the increasing number of prototypes does not increase the accuracy significantly. In fact, the increasing number of prototypes used can increase the computational time.