Forecasts for the Ecology and Fisheries Economy of Scottish herring and mackerel

Abstract

In this project, we aimed to develop an efficient supervised learning system foranalyzing second-hand sailboat prices in Hong Kong. The project was divided into three main steps: data cleaning and denoising, dimensionality reduction, and efficient supervised learning. In the first step, we successfully cleaned and denoised the data by filling the missing values using mean, mode. We also detected and removed 154 outliers and abnormal points through Q-Q diagram. The remaining data passed the normality test after dimensionless standardization, and we confirmed that they all conformed to a normal distribution. In the second step, we reduced the dimensionality of the data by combining Pearson correlation coefficient and the mRMR algorithm. We selected the top 6 features as the inputs for the supervised learning model. In the third step, we established a supervised learning system with the second-hand sailboat price as the top layer, quality of the sailboat, year of production, region, and volume as the middle layer, and the remaining small index features as the bottom layer. We used an SVM model with penalty conditions for the S-N layer and a random forest model with parameter adjustment for the S-P layer. The S-N model achieved an accuracy of over 85%, AUC of 0.93, while the R2 of the S-P model was greater than 0.87 and the RMSE was less than 0.7, indicating the model was well optimized. Overall, our project successfully established an efficient supervised learning system to analyze second-hand sailboat prices in Hong Kong, providing insights into the regional effect and enabling better decision-making in the sailboat market.