Most previous studies on forecasting shipping market conditions involved traditional time series methods using quantitative data or artificial intelligence (AI) methods. This study takes a different approach from traditional quantitative analysis methods and presents a method for forecasting shipping market conditions which utilizes unstructured and previously unused data (such as reports and articles). We applied a support vector machine (SVM)-based machine learning method to classify whether such data on shipping market conditions have a “positive” or “negative” impact on weekly forecast of shipping (container) market conditions. To assess the performance of the proposed technique, we conducted a test in which we compared its performance with that of a sentiment analysis method. According to the forecast test results with Lloyd's List container market conditions [Shanghai Containerized Freight Index (SCFI) (composite)], the sentiment analysis forecast success rate was 51.7%, and that of the SVM model was 58.6%. Using PR News container market conditions [SCFI (composite)], the forecast success rate was 37.55 and 74.3% respectively for the sentiment analysis and the SVM model. Thus, the SVM model demonstrated better forecasting performance than the existing sentiment analysis classification model
Ⅰ. Introduction
Ⅱ. SVM Method and Previous Reserch
Ⅲ. Container Market Analysis Data and Procedure
Ⅳ. Analysis Results
Ⅴ. Conclusion
References