Comparison of three LDA, PCA and ICA fast methods using fourteen data analysis algorithms to develop a risk assessment management model for export declarations to deal with illegal trade in Iran customs

Risk assessment is the main component of risk management, therefore, developing a suitable data analysis model is particularly important in customs. The purpose of this research is to use data mining techniques to develop an intelligent model for timely prediction of the risk level of export declarations in customs and as a result to prevent irreparable damages. Data mining techniques have been used in this research considering the data-oriented statistical population. The statistical data of the cross-border trade system of the Iranian customs is 698,781 data of the export declaration of the entire customs of the country of Iran for the year 2019-2020. Using Python programming language, feature reduction and effective feature extraction were performed after data preprocessing and preparation, with three methods of principal component analysis, linear differential analysis, and fast independent component analysis. Then for the predictive modelling of fourteen classification algorithms, three methods of principal component analysis (PCA), linear discriminant analysis (LDA) and fast independent component analysis (Fast ICA) were used and eighty percent of the training data were used. After training the models, forty-two different models were extracted. For testing, the obtained models were tested with twenty percent of the data. The test results of the models were compared with standard metrics to evaluate the efficiency of the models and the model obtained from the random forest algorithm with the fast independent component analysis method with three features was selected as the best model for predicting and determining the risk level of export declarations in customs.