SCFI Prediction Using SVR Model

 By using the regression algorithm of SVM, which is a representative algorithm of machine learning, I want to predict the recently soaring container freight index.

 

 Machine learning is one of the research fields of artificial intelligence, and it refers to the process of building a computer program to improve or predict an existing model or outcome through learning based on empirical data. In other words, it refers to finding regularities or patterns by using statistics and algorithms based on input data, and then building a specific model and obtaining the necessary answers.

The part used in this study is machine learning-supervised learning-regression.

---

1. Overview of SVM and SVR Models

 

 A support vector machine (SVM) is an algorithm that searches for a boundary during classification and classifies based on the boundary when new data is received. That is, the support vector can be said to be a vector that supports defining a boundary line between points displayed in a multidimensional space.

The goal of SVM is to divide the space and create flat borders (hyperplanes) on both sides. The distance between these boundary lines and the support vectors is called the margin. No matter how accurately the classification boundary can be set, if the margin is small, an error may occur in classifying new data.

---

2. Modeling

2.1.  Data extraction

Date	SCFI
01-Jan-2021	2,783.03
08-Jan-2021	2,870.34
15-Jan-2021	2,885.00
22-Jan-2021	2,868.95
29-Jan-2021	2,861.69
05-Feb-2021	2,884.61
12-Feb-2021	2,825.75
19-Feb-2021	2,875.93
26-Feb-2021	2,775.29

 The Composite Freight Index (SCFI) was extracted from Clarkson. To reflect the recent sharp rise, we intend to use 65 weekly data from January 1, 21, to April 1, 22.

 A scatterplot of the extracted data is drawn as above. Since the counting started in 2009 (starting with 1,000 points), the highest point was reached in January 22nd. It is true that the current level is very high, but it should be taken into account that it has been showing a downward trend for the last 10 weeks in a row. The red line is a regression graph analyzed through the SVR algorithm, and is a line connecting the above-mentioned margin.

 

2.2. Optimization

 

 For model optimization, I want to compare values before and after algorithm tuning (cost value change) and select a model with low prediction error. I tried to find the combination that finds the maximum margin under the cost range.

Type	RMSE
Tuned Before	101.9201
Tuned After	84.68043

 

Looking at the table above, it was confirmed that the prediction error (RMSE) value was low when the tuning process for optimization was performed. Therefore, we want to predict the future price using the tuned model.

 

 

3. Forecasts and Conclusions

Week	Forecast
15W	4209.991
16W	4076.193
17W	3942.394

 

 

 Based on the SCFI index as of April 1 (4,348.71), this is predicted up to three weeks into the future. According to the SVR model, it was said that freight rates will continue to decline until three weeks from now, but it seems to be the case to see how it will turn out.

 The SVM model (classification) summarized in this article has already been widely used in various classification fields such as language identification, document classification by subject, and detection of important events. The SVR model is also being used in various academic fields such as apartment building construction cost, wind speed, and drought prediction.

 

Thanks.