Financial Condition Prediction Using a Soft Voting Ensemble Model Based on Structured Financial Indicators and Unstructured News Data

Abstract

The rapid growth of capital market participation has increased the need for reliable analytical tools to assess the financial conditions of listed companies. Conventional approaches commonly rely on structured financial indicators and may not fully capture contextual information reflected in financial news. This study aims to evaluate the predictive potential of structured financial data and unstructured textual data using a soft voting ensemble framework. The structured dataset consists of 1,263 financial records collected from the reports of manufacturing companies listed on the Indonesia Stock Exchange, whereas the unstructured dataset contains 6,329 online financial news records related to Indonesian stocks and listed companies. The structured data were processed through missing-value handling, normalization, and class balancing, while the textual data were processed through cleaning, case folding, tokenization, filtering, stemming, and Term Frequency–Inverse Document Frequency (TF-IDF) feature extraction. The proposed ensemble model combines the probability outputs of six base classifiers: Decision Tree, Support Vector Machine, Multinomial Naive Bayes, Logistic Regression, Random Forest, and K-Nearest Neighbors. Experimental results show that the soft voting ensemble achieved an accuracy of 92.66% on the structured dataset and 98.53% on the unstructured dataset. These findings indicate that both financial indicators and textual information can provide valuable predictive signals for identifying company financial conditions. The contribution of this study lies in the comparative evaluation of two distinct data sources using a consistent ensemble-learning framework. However, the two datasets were evaluated independently rather than integrated into a single multimodal pipeline. Therefore, future research should develop a data-fusion mechanism to assess whether combining financial indicators and news-based sentiment can further improve predictive performance and strengthen decision support for investors and financial analysts.