Enthusiastic : International Journal of Applied Statistics and Data Science

Loan Approval Classification Using Ensemble Learning on Imbalanced Data

Rahmi Anadra — 2024-10-01

Loan processing is an important aspect of the financial industry, where the right decisions must be made to determine loan approval or rejection. However, the issue of default by loan applicants has become a significant concern for financial institutions. Hence, ensemble learning needs to be used with random forest and Extreme Gradient Boosting (XGBoost) algorithms. Unbalanced data are handled using the Synthetic Minority Over-sampling Technique (SMOTE). This research aimed to improve accuracy and precision in credit risk assessment to reduce human workload. Both algorithms used a dataset of 4,296 with 13 variables relevant to making loan approval decisions. The research process involved data exploration, data preprocessing, data sharing, model training, model evaluation with accuracy, sensitivity, specificity, and F1-score, model selection with 10-fold cross-validation, and important variables. The results showed that XGBoost with imbalanced data handling had the highest accuracy rate of 98.52% and a good balance between sensitivity of 98.83%, specificity of 98.01, and F1-score of 98.81%. The most important variables in determining loan approval are credit score, loan term, loan amount, and annual income.

Detection and Quantification of Glandular Trichomes (Bulbous) on Potato Plant Leaf Images Using Deep Learning

M. Fauzan Azhari — 2024-10-01

Potato plants have a very high nutritional value, making them widely cultivated in Indonesia. To ensure the cultivation of potatoes has good quality, many individuals, ranging from farmers to researchers and plant breeders, strive to explore and understand the characteristics of plant resistance sources, one of which is through the role of trichomes. Trichomes are fine hairs that coat the outer surface of plant leaves, serving as a physical barrier and regulating plant temperature. Identification and quantification of trichomes are commonly conducted manually by researchers, which consumes much time and is inefficient. Therefore, a system that can automatically detect and quantify trichomes is crucial to avoid manual identification and quantification, allowing these processes to be carried out more quickly. This study utilized a deep learning approach to train a model capable of detecting and quantifying trichome objects. The model architecture used was YOLOv8. From the training process, the resulting mean average precision (mAP) at a confidence threshold of 50 was 0.816, while the mAP at a confidence threshold of 90 was 0.38. This model is expected to assist experts or researchers in the field of agriculture in identifying trichomes, thereby optimizing crop yields.

Factor Influencing Delayed Completion in Mathematics Students at Nusa Cendant University: A Factor Analysis Approach

Elisabeth Brielin Sinu — 2024-10-30

This study aimed to identify and analyze factors contributing to the delay in the study period of students enrolled in the Mathematics Program at the Faculty of Science and Engineering (FST), Nusa Cendana University (UNDANA). The research employed a comprehensive analytical approach, starting with validity and reliability tests, followed by descriptive analysis, and culminating in factor analysis. Initially, 27 variables were considered; however, after conducting validity and reliability assessments, 18 variables were deemed suitable for further analysis. These 18 variables were subjected to factor analysis, revealing that they could be consolidated into four distinct factors, collectively accounting for 68.734% of the total variability observed among the students. The four identified factors influencing study delays are (1) student and supervisor commitment to completing the final project, (2) campus and peer support, (3) intelligence and discipline, and (4) motivation and relationships. Among these, the commitment of students and their supervisors to the timely completion of the final project emerged as the most dominant factor, demonstrating 43.417% of the total variance. The findings highlight the crucial role of both individual dedication and external support systems in ensuring timely academic progress, offering valuable insights for improving student outcomes in the Mathematics Program at UNDANA.

Statistical Perspective of Dengue Hemorrhagic Fever in West Java: Insights from Two-Way RE Model

Ghiffari Ahnaf Danarwindu — 2024-10-30

The Indonesian Ministry of Health has reported an alarming increase in Dengue Hemorrhagic Fever (DHF) cases, particularly in West Java Province. Given this trend, collaborative research and surveillance efforts are crucial to understanding and managing DHF cases in Indonesia. The panel data regression model in dengue fever cases will provide new insights into modeling. This research aimed to identify the most appropriate random effects model for estimating a dataset with four different variables. This study involved panel data variables on the effect of population density, percentage of poor people, percentage of households with access to clean water, and proper sanitation on DHF cases in West Java Province. This method emphasized selecting the best model from one-way and two-way Random Effects (RE) models and identifying what factors influenced the increase of DHF cases in West Java province. The best model obtained was a two-way RE Model with three significant variables. Based on the selected variables in the model, West Java Province needs to pay attention to the distribution of housing and economic activity in each district because population density is a crucial concern for the local government.

Indonesian Inflation Forecasting with Recurrent Neural Network Long Short-Term Memory (RNN-LSTM)

Hermansah — 2024-10-30

This study forecasted inflation in Indonesia using the Recurrent Neural Network Long Short-Term Memory (RNN-LSTM) model, ideal for nonlinear, complex time series data. It evaluated the effects of different activation functions, such as Logistic, Gompertz, and Hyperbolic Tangent (tanh); and weight update methods, such as Stochastic Gradient Descent (SGD) and Adaptive Gradient (AdaGrad) on RNN-LSTM performance. Monthly inflation data from January 2005 to December 2023 underwent preprocessing, including normalization and autoregressive lag-based input selection. Model accuracy was assessed with Root Mean Squared Error (RMSE) and Symmetric Mean Absolute Percentage Error (SMAPE). The findings indicated that the RNN-LSTM model with the logistic activation function and SGD optimization achieved the highest accuracy, outperforming traditional models such as Exponential Smoothing (ETS), Autoregressive Integrated Moving Average (ARIMA), Feedforward Neural Network (FFNN), and Recurrent Neural Network (RNN). Additionally, optimal learning rate and epoch values were identified, enhancing model stability and precision. In conclusion, the study confirms that the RNN-LSTM model is effective for inflation forecasting when optimized with specific activation functions and optimization methods. It recommends further exploration of neuron configurations and alternative models, such as the Gated Recurrent Unit (GRU), to improve forecast accuracy.

Modeling the Number of Foreign Tourist Visits to Indonesia in 2020 Using GWPR Method

Muhammad Zidni Subarkah — 2024-10-30

In December 2020, the number of foreign tourists visiting Indonesia experienced a sharp decline of 88.08% compared to the number of visits in December 2019. However, compared to the previous month, November 2020, this number increased by 13.58%. Modeling the number of foreign tourists visiting Indonesia in 2020 using the Geographically Weighted Poisson Regression (GWPR) method is needed to elaborate on the Indonesian government’s policy decisions, especially in the tourism sector. The results showed that the GWPR model with the Kernel fixed Gaussian weighted function had an AIC value of 1,521,240.873, deviance of 1,521,196.695, and deviance-R2 of 0.741 or 74.1%. This model produced two different clusters of characteristics of foreign tourists’ country of origin based on the variable’s significance. Cluster one consisted of Finland and Qatar and the rest were in cluster two. The characteristics of cluster two were influenced by the rupiah exchange rate variable, short stay visa free (Bebas Visa Kunjungan Singkat, BVKS), Consumer Price Index (CPI), economic growth, total imports, and the distance of CGK to the international airport. Meanwhile, cluster one had almost the same characteristics as cluster two but was not influenced by the BVKS factor variables.

Evaluating Creative Therapy Effectiveness on Children with Special Needs through Robust Clustering Techniques

Rahmadi Yotenka — 2024-10-30

The study examined the progress of Children with Special Needs (CWSN) in the Center for Students with Special Needs (Pusat Layanan Peserta Didik Berkebutuhan Khusus, PLPDBK) Semarang through creative therapy methods. Based on the primary data collected from the observation of 56 children over eight sessions of therapy. The study employed the Robust Clustering Using Links (ROCK) clustering algorithm to evaluate children’s social interaction and behavior development, fine motor skills, and cognitive capabilities. The clustering process revealed four distinct types of CWSN that, for the most part, were between the ages of 6 and 10 years old. The study found that although the stability of these development features was often seen, there was a possibility for improvements in certain categories. The study highlighted the potential of targeted interventions and modern treatments that regularly elevate children to “5” or the “very good” developmental category during the vital age range of 6 to 10 years. These findings call for greater inclusion in educational policy and therapies that can be designed to accommodate the various needs of children.