• Journal of Korean Society of Industrial and Systems Engineering
• /
• v.44 no.2
• /
• pp.15-23
• /
• 2021

Most of the open-source decision tree algorithms are based on three splitting criteria (Entropy, Gini Index, and Gain Ratio). Therefore, the advantages and disadvantages of these three popular algorithms need to be studied more thoroughly. Comparisons of the three algorithms were mainly performed with respect to the predictive performance. In this work, we conducted a comparative experiment on the splitting criteria of three decision trees, focusing on their interpretability. Depth, homogeneity, coverage, lift, and stability were used as indicators for measuring interpretability. To measure the stability of decision trees, we present a measure of the stability of the root node and the stability of the dominating rules based on a measure of the similarity of trees. Based on 10 data collected from UCI and Kaggle, we compare the interpretability of DT (Decision Tree) algorithms based on three splitting criteria. The results show that the GR (Gain Ratio) branch-based DT algorithm performs well in terms of lift and homogeneity, while the GINI (Gini Index) and ENT (Entropy) branch-based DT algorithms performs well in terms of coverage. With respect to stability, considering both the similarity of the dominating rule or the similarity of the root node, the DT algorithm according to the ENT splitting criterion shows the best results.

• Proceedings of the KSRS Conference
• /
• 2008.10a
• /
• pp.53-56
• /
• 2008

Generally, the quality of satellite images is expressed by GSD (Ground Sample Distance), MTF (Modulation Transfer Function) and SNR (Signal to Noise Ratio). However, these factors are technology-oriented and do not explain interpretability of satellite images. We need a standardized index which shows standard of interpretability. In this study, we estimated NIIRS (National Imagery Interpretability Rating Scale) through the GIQE (General Image Quality Equation) which is able to judge image interpretability with the standardized index. Traditionally, NIIRS has been determined manually by specialized image analysts. We used the GIQE in order to reduce inefficiency and high costs cause by manual interpretation and to produce accurate NIIRS. For monitoring image degradation, we estimated GIQE physical parameters from image analysis and carried out time series analysis about the quality of the KOMPSAT-1 images. On all of the tests, we were able to identify the image degradation due to the changing time. This indicates that NIIRS derived from GIQE will be used for image degradation indicator.

• Journal of Intelligence and Information Systems
• /
• v.15 no.4
• /
• pp.123-140
• /
• 2009

Fuzzy neural network is an integrated model of artificial neural network and fuzzy system and it has been successfully applied in control and forecasting area. Recently ANFIS(Adaptive Network-based Fuzzy Inference System) has been noticed widely among various fuzzy neural network models because of outstanding performance of control and forecasting accuracy. ANFIS has capability to refine its fuzzy rules interactively with human expert. In particular, when we use initial rule structure for machine learning which is generated from human expert, it is highly probable to reach global optimum solution as well as shorten time to convergence. We propose metrics to evaluate interpretability of generated rules as a means of acquiring domain knowledge and compare level of interpretability of ANFIS fuzzy rules to those of C5.0 classification rules. The proposed metrics also can be used to evaluate capability of rule generation for the various machine learning methods.

• Structural Engineering and Mechanics
• /
• v.84 no.5
• /
• pp.605-618
• /
• 2022

Glass fiber reinforced polymer (GFRP) elastic gridshells consist of long continuous GFRP tubes that form elastic deformations. In this paper, a method for the form-finding of gridshell structures is presented based on the interpretable machine learning (ML) approaches. A comparative study is conducted on several ML algorithms, including support vector regression (SVR), K-nearest neighbors (KNN), decision tree (DT), random forest (RF), AdaBoost, XGBoost, category boosting (CatBoost), and light gradient boosting machine (LightGBM). A numerical example is presented using a standard double-hump gridshell considering two characteristics of deformation as objective functions. The combination of the grid search approach and k-fold cross-validation (CV) is implemented for fine-tuning the parameters of ML models. The results of the comparative study indicate that the LightGBM model presents the highest prediction accuracy. Finally, interpretable ML approaches, including Shapely additive explanations (SHAP), partial dependence plot (PDP), and accumulated local effects (ALE), are applied to explain the predictions of the ML model since it is essential to understand the effect of various values of input parameters on objective functions. As a result of interpretability approaches, an optimum gridshell structure is obtained and new opportunities are verified for form-finding investigation of GFRP elastic gridshells during lifting construction.

• Korean Journal of Remote Sensing
• /
• v.27 no.5
• /
• pp.587-599
• /
• 2011

As one measure of image interpretability, NIIRS(National Imagery Interpretability Rating Scale) has been used. Unlike MTF(Modulation Transfer Function), SNR(Signal to Noise Ratio), and GSD(Ground Sampling Distance), NIIRS can describe the quality of overall image at user's perspective. NIIRS is observed with human observation directly or estimated by edge analysis. For edge analysis specially manufactured artificial target is used commonly. This target, formed with a tarp of black and white patterns, is deployed on the ground and imaged by the satellite. Due to this, the artificial target-based method needs a big expense and can not be performed often. In this paper, we propose a new edge analysis method that enables to estimate NIIRS accurately. In this method, natural targets available in the image are used and characteristics of the target are considered. For assessment of the algorithm, various experiments were carried out. The results showed that our algorithm can be used as an alternative to the artificial target-based method.

• Journal of Korea Multimedia Society
• /
• v.19 no.4
• /
• pp.725-735
• /
• 2016

High-resolution satellite images are used in the fields of mapping, natural disaster forecasting, agriculture, ocean-based industries, infrastructure, and environment, and there is a progressive increase in the development and demand for the applications of high-resolution satellite images. Users of the satellite images desire accurate quality of the provided satellite images. Moreover, the distinguishability of each image captured by an actual satellite varies according to the atmospheric environment and solar angle at the captured region, the satellite velocity and capture angle, and the system noise. Hence , NIIRS must be measured for all captured images. There is a significant deficiency in professional human resources and time resources available to measure the NIIRS of few hundred images that are transmitted daily. Currently, NIIRS is measured every few months or even few years to assess the aging of the satellite as well as to verify and calibrate it [3]. Therefore, we develop an algorithm that can measure the national image interpretability rating scales (NIIRS) of a typical satellite image rather than an artificial target satellite image, in order to automatically assess its quality. In this study, the criteria for automatic edge region extraction are derived based on the previous works on manual edge region extraction [4][5], and consequently, we propose an algorithm that can extract the edge region. Moreover, RER and H are calculated from the extracted edge region for automatic edge region extraction. The average NIIRS value was measured to be 3.6342±0.15321 (2 standard deviations) from the automatic measurement experiment on a typical satellite image, which is similar to the result extracted from the artificial target.

• International Journal of Control, Automation, and Systems
• /
• v.5 no.4
• /
• pp.444-455
• /
• 2007

An approach to construct multiple interpretable and precise fuzzy systems based on the Pareto Multi-objective Cooperative Coevolutionary Algorithm (PMOCCA) is proposed in this paper. First, a modified fuzzy clustering algorithm is used to construct antecedents of fuzzy system, and consequents are identified separately to reduce computational burden. Then, the PMOCCA and the interpretability-driven simplification techniques are executed to optimize the initial fuzzy system with three objectives: the precision performance, the number of fuzzy rules and the number of fuzzy sets; thus both the precision and the interpretability of the fuzzy systems are improved. In order to select the best individuals from each species, we generalize the NSGA-II algorithm from one species to multi-species, and propose a new non-dominated sorting technique and collaboration mechanism for cooperative coevolutionary algorithm. Finally, the proposed approach is applied to two benchmark problems, and the results show its validity.

• International journal of advanced smart convergence
• /
• v.12 no.1
• /
• pp.53-58
• /
• 2023

With the fast development of artificial intelligence day by day, users are demanding explanations about the results of algorithms and want to know what parameters influence the results. In this paper, we propose a model for bankruptcy prediction with interpretability using the SHAP framework. SHAP (SHAPley Additive exPlanations) is framework that gives a visualized result that can be used for explanation and interpretation of machine learning models. As a result, we can describe which features are important for the result of our deep learning model. SHAP framework Force plot result gives us top features which are mainly reflecting overall model score. Even though Fully Connected Neural Networks are a "black box" model, Shapley values help us to alleviate the "black box" problem. FCNNs perform well with complex dataset with more than 60 financial ratios. Combined with SHAP framework, we create an effective model with understandable interpretation. Bankruptcy is a rare event, then we avoid imbalanced dataset problem with the help of SMOTE. SMOTE is one of the oversampling technique that resulting synthetic samples are generated for the minority class. It uses K-nearest neighbors algorithm for line connecting method in order to producing examples. We expect our model results assist financial analysts who are interested in forecasting bankruptcy prediction of companies in detail.

• Journal of Korean Society for Geospatial Information Science
• /
• v.8 no.2 s.16
• /
• pp.61-69
• /
• 2000

Until now, the extraction of information on geographic features and the compilation of maps from satellite imagery has had many limitations because of its lower resolution compared to aerial photos to the recent. However, it is expected that the availability of high resolution satellite imagery whose spatial resolution is about 1m will reduce such limitations. Currently, a compilation of national-wide digital base maps is going on to construct the National Geographic Information Systems in Korea. It will be used for many application field of the social welfare. Therefore, in this study, we suggest that satellite imagery can help it and we have experimented on the possibility of detecting and interpreting geographic data using satellite imagery of various spatial resolutions. The interpretability and detectability of 46 features in 6 categories was experimented with 6 kinds of images of different resolutions. As a subsequent procedure, we have performed the fractal analysis for a quality test of the texture information. Through the fractal analysis, we could show that texture information and probability of discrimination increases as the spatial resolution of the image increases. Based on the results of this experiment, we could suggest the possibility of the renewal and construction of the National-wide Geographic Information Systems database using satellite imagery, as well as of examining appropriate spatial resolutions for objects of interest.

• Journal of Intelligence and Information Systems
• /
• v.29 no.2
• /
• pp.241-265
• /
• 2023

A corporate insolvency prediction model serves as a vital tool for objectively monitoring the financial condition of companies. It enables timely warnings, facilitates responsive actions, and supports the formulation of effective management strategies to mitigate bankruptcy risks and enhance performance. Investors and financial institutions utilize default prediction models to minimize financial losses. As the interest in utilizing artificial intelligence (AI) technology for corporate insolvency prediction grows, extensive research has been conducted in this domain. However, there is an increasing demand for explainable AI models in corporate insolvency prediction, emphasizing interpretability and reliability. The SHAP (SHapley Additive exPlanations) technique has gained significant popularity and has demonstrated strong performance in various applications. Nonetheless, it has limitations such as computational cost, processing time, and scalability concerns based on the number of variables. This study introduces a novel approach to variable selection that reduces the number of variables by averaging SHAP values from bootstrapped data subsets instead of using the entire dataset. This technique aims to improve computational efficiency while maintaining excellent predictive performance. To obtain classification results, we aim to train random forest, XGBoost, and C5.0 models using carefully selected variables with high interpretability. The classification accuracy of the ensemble model, generated through soft voting as the goal of high-performance model design, is compared with the individual models. The study leverages data from 1,698 Korean light industrial companies and employs bootstrapping to create distinct data groups. Logistic Regression is employed to calculate SHAP values for each data group, and their averages are computed to derive the final SHAP values. The proposed model enhances interpretability and aims to achieve superior predictive performance.