Abstract. Advances in deep learning have significantly impacted various sectors, including agriculture, by improving efficiency and sustainability. Computer vision-based plant disease detection that utilizes deep learning models such as Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) has replaced manual methods that require specialized skills. Vision Transformers (ViTs) with self-attention mechanisms excel at recognizing complex patterns in images of diseased plants, enabling early detection and more effective management. This study conducted a systematic literature review to evaluate the use of feature fusion-based transformer algorithms in plant disease detection and to identify current research trends, challenges, and future opportunities. The results show that ViTs can improve accuracy and efficiency in plant disease diagnosis, particularly under diverse environmental conditions. However, the adoption of this technology faces obstacles such as the need for large annotated datasets, significant computational costs, and variability in field conditions that can affect model performance. This research highlights the need for further innovation to address these challenges and expand the accessibility and reliability of deep learning-based plant disease detection for the global agricultural sector.

Abstract. This paper analyzed the feature selection for reducing the number of input variables when developing a predictive model. Boruta Algorithm is using in this paper as a wrapper around a Random Forest classification algorithm. Boruta algorithm is one of the algorithms used to determine the significant variables (feature selection) in a classification model in the machine learning approach, as supervised learning. Our results show that on the German Credit Data from the UCI Machine Learning with 20 variables, feature selection using Boruta Algorithm with Python Programming obtains 4 significant features.

Abstract. This paper addressed the fuzzy logic and integrated Dijkstra’s algorithm to determine an efficient path for travel routes planning. Dijkstra’s algorithm is the same as Breadth-First Search (BFS) algorithm, i.e. queuing principle, but the queue in Dijkstra’s Algorithm is a priority queue. The research methodology consists of data collection, clustering, fuzzy logic modelling, implementation of the Dijkstra algorithm, and testing the results. The collected data in this paper is the distance and travel time. Data clustering is needed to classify time data based on congestion level because travel time data is dynamic. The results of the clustering become a reference for selecting the time in collecting travel time data. The time represents the current state of the traffic. Fuzzy logic modelling is done after all the required data has been collected. This paper uses the Tsukamoto fuzzy method because this method has tolerance for data and is very flexible. The data is developed into a mathematical model to produce a crisp output (i.e. travel weight). The crisp output is then processed with Dijkstra’s algorithm to produce a solution that solves the problem in this paper. The final result of this research is that fuzzy logic and integrated Dijkstra’s algorithm is ideal for efficient pathfinding because it always gives the most efficient results from all possible paths available with the existing problem constraints.

Abstract We compare performance of six single classifiers trained on German credit dataset, an imbalanced dataset of 1000 instances with binary-valued dependent variable. To improve the performance, we consider resampling the dataset and ensembling the classifiers. The benchmarks are taken from the best performance among six considered classifiers. Resampling the dataset includes oversampling and undersampling. The performance of ensemble classifiers are then analyzed and examined. The experimental results provide three benchmarks, i.e. SVM trained on plain dataset, NB trained on plain dataset, and SVM trained on undersampled dataset. Furthermore, ensemble of kNN, LDA and SVM outperforms the first benchmark for all metrics used in this research, i.e. recall 92.71%, precision 79.14%, F1 84.73%, AUC 79.96%, and accuracy 76.88%. The ensemble of LR, SVM and NB and the ensemble of LDA, SVM, and NB outperforms the second and third benchmark, respectively.

Abstract: To accelerate the financial services, microfinance requires tools and technologies to provide an automated dynamic credit decision which leads to an accountable and efficient system. Considering a case on loan disbursement in the micro-business sector, this study presents a very comprehensive innovation, namely automated credit intelligence learning system (ACILES) which consists of dynamic credit scoring and optimal dynamic credit pricing: derived from tenor, rate, installment and plafond (TRIP). While credit pricing is obtained from the profit based pricing and simulation process, the credit scoring is developed by modelling not only the borrower’s profile, but also psychometric analysis of the perception of borrower and surveyor via item response model which is combined with multivariate adaptive regression splines (MARS) model and structural equation modelling (SEM), respectively. By performing the experiment, it is clearly proved that ACILES can be implemented in order to augment microfinance business capacity.

Abstract. Credit scoring is the main process of credit transactions in assessing the credit risk of credit applicants. Unfortunately, in practice, its implementation only stops to the credit approvals. In this research, we utilize credit scores to generate the customized credit prices. We believe that each person has their own credit risk so that they will get different credit prices depend on their individual credit risk. This credit risk-based pricing is optimized by reinforcement learning approaches to represent the dynamic solution related to the updated credit historical data. There are several variables considered in the profit optimization model such as credit scores, tenor, credit prices (or rate for credit applicants), and plafond. We implement this solution to the random generated credit data.