Data mining is an approach to discover knowledge from large data. Pollutant forecasting is an imp... more Data mining is an approach to discover knowledge from large data. Pollutant forecasting is an important problem in the environmental sciences. This paper tries to use data mining methods to forecast fine particles (PM2.5) concentration level in Hong Kong Central, which is a famous business centre in Asia. There are several classification algorithms available in data mining, such as Artificial Neural Network (ANN) and Support Vector Machine (SVM). ANN and SVM are both machine learning algorithm used in variant area. This paper builds PM2.5 concentration level predictive models based on ANN and SVM by using R packages. The data set includes 2008-2011 period meteorological data and PM2.5 data. The PM2.5 concentration is divided into 2 levels: low and high. The critical point is 40μg/m3 (24 hours mean), which is based on the standard of US Environmental Protection Agency (EPA). The parameters of both models are selected by multiple cross validation. According to 100 times 10-fold cross validation, the testing accuracy of SVM is around 0.803~0.820, which is much better than ANN whose accuracy is around 0.746~0.793.
Data mining is an approach to discover knowledge from large data. Pollutant forecasting is an imp... more Data mining is an approach to discover knowledge from large data. Pollutant forecasting is an important problem in the environmental sciences. This paper tries to use data mining methods to forecast fine particles (PM2.5) concentration level in a new town of Hong Kong rural area. There are several classification algorithms available in data mining, such as Artificial Neural Network (ANN), Boosting, k-Nearest Neighbours (k-NN), and so forth. All of them are popular machine learning algorithms in various data mining areas, which including environmental data mining, educational data mining, financial data mining, etc. This paper builds PM2.5 concentration level predictive models based on ANN, Boosting (i.e. AdaBoost.M1), k-NN by using R packages. The data set includes 2009 to 2011 period meteorological data and PM2.5 data. The PM2.5 concentration is divided into 2 levels: low and high. The critical point is 25μg/m3 (24 hours mean. The parameters of both models are selected by multiple cross validation. According to 100 replications of 10-fold cross validation, the testing accuracy of AdaBoost.M1 is around 0.846~0.868, which is the best result among three algorithms in this paper.
Pollutant forecasting is an important problem in the environmental sciences. Data mining is an ap... more Pollutant forecasting is an important problem in the environmental sciences. Data mining is an approach to discover knowledge from large data. This paper tries to use data mining methods to forecast concentration level, which is an important air pollutant. There are several tree-based classification algorithms available in data mining, such as CART, C4.5, Random Forest (RF) and C5.0. RF and C5.0 are popular ensemble methods, which are, RF builds on CART with Bagging and C5.0 builds on C4.5 with Boosting, respectively. This paper builds concentration level predictive models based on RF and C5.0 by using R packages. The data set includes 2000-2011 period data in a new town of Hong Kong. The concentration is divided into 2 levels, the critical points is 25μg/m3 (24 hours mean). According to 100 times 10-fold cross validation, the best testing accuracy is from RF model, which is around 0.845~0.854.
PM2.5 refers to particulate matter whose size is 2.5 micrometers or smaller. It is one of the mos... more PM2.5 refers to particulate matter whose size is 2.5 micrometers or smaller. It is one of the most serious pollutants in some Asian countries, especially in big cities. Hence finding predictive model of PM2.5 is important in environmental science and related areas. This thesis uses data mining algorithms to build predictive model for next day’s PM2.5 concentration level, based on the standard of United States Environmental Protection Agency. One of the challenges in this thesis is that our PM2.5 predictive model is a multi-level task which means the response variable has three levels. This is much more complicated than binary response tasks in pre-diction process. Generally speaking, logistic regression (LR) is the classical statistical method for classification task. Additionally, there are also several classification algorithms available in data analysis and statistics, such as classification and regression tree (CART), adaptive boosting (AdaBoost), bootstrap aggregating (bagging) and random forest (RF). All of them are popular machine learning algorithms in various data mining areas, which include environmental data mining, educational data mining, financial data mining and others. We will start with LR and some tree-based algorithms. Another contribution in this thesis is that we introduce a new method which is named majority voting logistic regression (mvLR). It can be considered as an integrated tool of traditional LR and majority voting method which are used in bagging and random forest. We will use these methods to train and test PM2.5 data set in order to select the proper parameters. Lastly, all of the tools (with tuned parameters) will be compared with each other by using multiple replications of 10-fold cross-validation (10-fold CV). Based on the me-dian accuracy (ACC) result, RF and mvLR perform better than others, where ACC are 0.774 and 0.769, respectively. Based on the median area under ROC curve (AUC) result, LR and RF obtain the higher performance than others, where AUC are 0.866 and 0.864, respectively.
This document is the study notes of “Introduction to MATRIX ALGEBRA” which was written by Autar K... more This document is the study notes of “Introduction to MATRIX ALGEBRA” which was written by Autar K. Kaw, University of South Florida.
This document is the solution manual of “MOOCULUS-2: Sequences and Series” which was provided by ... more This document is the solution manual of “MOOCULUS-2: Sequences and Series” which was provided by Coursera in 2014.
This document is a personal solution manual of BerkeleyX Stat2.3x “Introduction to Statistics: In... more This document is a personal solution manual of BerkeleyX Stat2.3x “Introduction to Statistics: Inference” which was provided by edX in 2014.
This document is a personal solution manual of BerkeleyX Stat2.2x “Introduction to Statistics: Pr... more This document is a personal solution manual of BerkeleyX Stat2.2x “Introduction to Statistics: Probability” which was provided by edX in 2014.
Adaboost is a powerful algorithm for predicting models. However, a major disadvantage is that Ada... more Adaboost is a powerful algorithm for predicting models. However, a major disadvantage is that Adaboost may lead to over-fit in the presence of noise. proved that the training error of the ensemble is bounded by the following expression:
This article tries to explain why growing a tree can reduce the error rate. In other words, is th... more This article tries to explain why growing a tree can reduce the error rate. In other words, is that better for splitting a node into its sub-nodes?
Data mining is an approach to discover knowledge from large data. Pollutant forecasting is an imp... more Data mining is an approach to discover knowledge from large data. Pollutant forecasting is an important problem in the environmental sciences. This paper tries to use data mining methods to forecast fine particles (PM2.5) concentration level in Hong Kong Central, which is a famous business centre in Asia. There are several classification algorithms available in data mining, such as Artificial Neural Network (ANN) and Support Vector Machine (SVM). ANN and SVM are both machine learning algorithm used in variant area. This paper builds PM2.5 concentration level predictive models based on ANN and SVM by using R packages. The data set includes 2008-2011 period meteorological data and PM2.5 data. The PM2.5 concentration is divided into 2 levels: low and high. The critical point is 40μg/m3 (24 hours mean), which is based on the standard of US Environmental Protection Agency (EPA). The parameters of both models are selected by multiple cross validation. According to 100 times 10-fold cross validation, the testing accuracy of SVM is around 0.803~0.820, which is much better than ANN whose accuracy is around 0.746~0.793.
Data mining is an approach to discover knowledge from large data. Pollutant forecasting is an imp... more Data mining is an approach to discover knowledge from large data. Pollutant forecasting is an important problem in the environmental sciences. This paper tries to use data mining methods to forecast fine particles (PM2.5) concentration level in a new town of Hong Kong rural area. There are several classification algorithms available in data mining, such as Artificial Neural Network (ANN), Boosting, k-Nearest Neighbours (k-NN), and so forth. All of them are popular machine learning algorithms in various data mining areas, which including environmental data mining, educational data mining, financial data mining, etc. This paper builds PM2.5 concentration level predictive models based on ANN, Boosting (i.e. AdaBoost.M1), k-NN by using R packages. The data set includes 2009 to 2011 period meteorological data and PM2.5 data. The PM2.5 concentration is divided into 2 levels: low and high. The critical point is 25μg/m3 (24 hours mean. The parameters of both models are selected by multiple cross validation. According to 100 replications of 10-fold cross validation, the testing accuracy of AdaBoost.M1 is around 0.846~0.868, which is the best result among three algorithms in this paper.
Pollutant forecasting is an important problem in the environmental sciences. Data mining is an ap... more Pollutant forecasting is an important problem in the environmental sciences. Data mining is an approach to discover knowledge from large data. This paper tries to use data mining methods to forecast concentration level, which is an important air pollutant. There are several tree-based classification algorithms available in data mining, such as CART, C4.5, Random Forest (RF) and C5.0. RF and C5.0 are popular ensemble methods, which are, RF builds on CART with Bagging and C5.0 builds on C4.5 with Boosting, respectively. This paper builds concentration level predictive models based on RF and C5.0 by using R packages. The data set includes 2000-2011 period data in a new town of Hong Kong. The concentration is divided into 2 levels, the critical points is 25μg/m3 (24 hours mean). According to 100 times 10-fold cross validation, the best testing accuracy is from RF model, which is around 0.845~0.854.
PM2.5 refers to particulate matter whose size is 2.5 micrometers or smaller. It is one of the mos... more PM2.5 refers to particulate matter whose size is 2.5 micrometers or smaller. It is one of the most serious pollutants in some Asian countries, especially in big cities. Hence finding predictive model of PM2.5 is important in environmental science and related areas. This thesis uses data mining algorithms to build predictive model for next day’s PM2.5 concentration level, based on the standard of United States Environmental Protection Agency. One of the challenges in this thesis is that our PM2.5 predictive model is a multi-level task which means the response variable has three levels. This is much more complicated than binary response tasks in pre-diction process. Generally speaking, logistic regression (LR) is the classical statistical method for classification task. Additionally, there are also several classification algorithms available in data analysis and statistics, such as classification and regression tree (CART), adaptive boosting (AdaBoost), bootstrap aggregating (bagging) and random forest (RF). All of them are popular machine learning algorithms in various data mining areas, which include environmental data mining, educational data mining, financial data mining and others. We will start with LR and some tree-based algorithms. Another contribution in this thesis is that we introduce a new method which is named majority voting logistic regression (mvLR). It can be considered as an integrated tool of traditional LR and majority voting method which are used in bagging and random forest. We will use these methods to train and test PM2.5 data set in order to select the proper parameters. Lastly, all of the tools (with tuned parameters) will be compared with each other by using multiple replications of 10-fold cross-validation (10-fold CV). Based on the me-dian accuracy (ACC) result, RF and mvLR perform better than others, where ACC are 0.774 and 0.769, respectively. Based on the median area under ROC curve (AUC) result, LR and RF obtain the higher performance than others, where AUC are 0.866 and 0.864, respectively.
This document is the study notes of “Introduction to MATRIX ALGEBRA” which was written by Autar K... more This document is the study notes of “Introduction to MATRIX ALGEBRA” which was written by Autar K. Kaw, University of South Florida.
This document is the solution manual of “MOOCULUS-2: Sequences and Series” which was provided by ... more This document is the solution manual of “MOOCULUS-2: Sequences and Series” which was provided by Coursera in 2014.
This document is a personal solution manual of BerkeleyX Stat2.3x “Introduction to Statistics: In... more This document is a personal solution manual of BerkeleyX Stat2.3x “Introduction to Statistics: Inference” which was provided by edX in 2014.
This document is a personal solution manual of BerkeleyX Stat2.2x “Introduction to Statistics: Pr... more This document is a personal solution manual of BerkeleyX Stat2.2x “Introduction to Statistics: Probability” which was provided by edX in 2014.
Adaboost is a powerful algorithm for predicting models. However, a major disadvantage is that Ada... more Adaboost is a powerful algorithm for predicting models. However, a major disadvantage is that Adaboost may lead to over-fit in the presence of noise. proved that the training error of the ensemble is bounded by the following expression:
This article tries to explain why growing a tree can reduce the error rate. In other words, is th... more This article tries to explain why growing a tree can reduce the error rate. In other words, is that better for splitting a node into its sub-nodes?
Uploads
Papers by Yin ZHAO
Thesis by Yin ZHAO
Notes by Yin ZHAO