Conference Agenda

In real-world classification problems, imbalanced data is often encountered, characterized by significantly different sample sizes among various classes for binary class and multiple class cases. In this study, machine learning algorithms are implemented after feature selection on imbalanced data sets. At first, resampling methods such as SMOTE, and Tomek Link are carried out to solve the class imbalance problem. Second, metaheuristic algorithms, namely Particle Swarm Optimization (PSO), Artificial Bee Colony (ABC), etc. are used for feature selection to determine relevant features. Finally, machine learning algorithms like K-nearest neighbor, Support Vector Machine, Decision Tree, Random Forest, and Logistic Regression are performed to classification aim. Comparative analysis is realized via different performance metrics, accuracy, precision, recall, F-score, and Matthew correlation coefficient to evaluate the performances of the mentioned algorithms for imbalanced data sets.

As a large number of credit scoring models are built on a known set of data (training data) collected in the past or from other regions/domains, a prerequisite for applying these models to new instances is that the data of the new instances are comparable to the training data set. The comparability between the new instances and the training dataset also has a strong impact on the performance of credit scoring models. However, most studies have focused on the methods or algorithms for model construction, there is a lack of research on the impact of the time lag of the training data on the accuracy of credit scoring models. This study aims to fill this gap by investigating how the time lag between training and test data sets affects the accuracy of credit scoring models. We collected 13 years of financial data from Japanese regional banks for the period 2010-2022. We used each year's data for 2010-2021 as training data to construct credit scoring models using support vector machines (SVMs). We then applied the models to predict the credit scores of banks for the following years and confirmed the accuracy of the models. It was clarified that (1) the accuracy of the models decreased as the time lag between training and test data increased; (2) to achieve an accuracy of more than 90%, it is necessary to construct a credit scoring model using data from the previous year.

This paper explores how Artificial Intelligence (AI) can revolutionize strategic foresight enhancing its capabilities and guiding decision-makers through an ever-changing landscape to navigate uncertainties through the systematic exploration of potential futures. The unparalleled pace of change encountered today necessitates robust foresight capabilities to enable businesses to navigate uncertainty. Traditional foresight techniques, while valuable, do not consider the vast quantities of available data or attempt to model complex causal relationships.

The paper discusses the potential of emerging deep learning AI techniques, such as Causal and Generative AI, to enhance traditional foresight approaches by bridging the gap between quantitative and qualitative foresight methods, addressing challenges such as expert bias and data reliability. Causal AI harnesses the power of causal inference to model complex causal relationships and mitigate observational bias in data, equipping decision-makers with robust tools for scenario building. Generative AI and LLMs on the other hand, offer promising capabilities for automated horizon scanning and scenario generation, although the need for human oversight remains Ultimately, AI could empower individuals and organizations to not only anticipate the future but actively shape it.

Key Words: Causal AI, Strategic Foresight, Bias Mitigation, AI, Artificial Intelligence, LLMs