31250 Ass3

Content

1. A description of the data mining problem;

2. The data preprocessing and transformations you did (if any);

3. How you went about solving the problem.

4. Classification techniques used and summary of the results and parameter settings.

5. The best classifier that you selected – the type, its performance, how it solved the problem (if it makes sense for that type of classifier), and reasons for selecting it.

6. Kaggle Submission Score

1. A description of the data mining problem.

Assessment 3 was tasked with building a classifier to classify “AIRLINE-NAME” that could be used in the data exploration process based on Assessment 2. In this task, but Knime is also a good fit for the visual programming software of this report, so I chose Knime for data mining and prediction. Data mining can not only help analysts understand data, but also help analysts make predictions and judgments through visual analysis and data mining results.

This task provides three data sets: FlightDataset. CSV, UnknownDataset. CSV and Kaggle-Submission-Random- sample.csv. The “training data” data set will be used to predict the data contained in the “Airline_Name” column. After classifying the data set, the prediction result will become the unknown data set of the prediction column, and the generated file format is the same as that of the kaggle-submitt-random-sample.csv data set.

The task requires the use of a variety of methods to predict the data and determine the maximum prediction accuracy of the method, and I will use several methods to predict.

2. The data preprocessing and transformations you did (if any);

First of all, use the CSV Reader node to import the CSV format file we want to use. Then use Rule Engine to achieve the conversion we need. The third step is the “Number to String” node to convert the number type to the string type. The picture above is what I set, I converted the data of Airline Name to string type, because the learner only deals with columns of string type.

I set the training set to 70%, the test set to 30%, and choose stratified sampling and random seed. Setting the random seed will make the sequence number the same every time, and will not cause the samples to be different, and the results of each validation will not be Change.

4. Classification techniques used and summary of the results and parameter settings.

Decision tree induction

A decision tree is a tree structure, each internal node represents an attribute test, each branch represents a test output, and each leaf node represents a category, which is a graphical method through probability analysis. In the above diagram, it shows how the decision tree classifier works and outputs the results to the scorer node and the ROC curve node.

The above three pictures are the configuration of the decision tree learner. Because the goal is to predict the Airline Name, I chose the Airline Name of the class column in the pruning method. I chose MDL because trying to use MDL can effectively improve the prediction accuracy. Then I set the number of threads to 2 and changed the minimum records per node to 8 to further improve the accuracy. The configuration of the decision tree predictor, changing the name of the prediction column to something that matches the task conditions, and then rechecking the box with the other columns of the normalized class distribution, it shows the normalized distribution for each prediction. and the ROC Curve node, I chose P (Airline_Name), which represents the predicted value of “Airline_Name”. Shows the results of equally split nodes, the accuracy of the decision tree model, and the accuracy of the error prediction.

Naive Bayes Predictor

This is a classification technique based on Bayes’ theorem, assuming independence among predictors. In short, a naive Bayesian classifier assumes that the presence of a particular feature in a class is independent of the presence of any other feature. Bayesian models are easy to build and are especially useful for very large datasets. It is well known that in addition to simplicity, Bayesian outperforms even very complex classification methods.

Random Forest

The Random Forest classifier used for classification is shown. Random Forest is composed of multiple decision trees, and the establishment of each tree relies on independent trees extracted independently. M samples are randomly selected from the original training sample set, and then new training samples are generated in the training decision tree set to form a random forest, and finally countless decision trees are generated. Random forest does not require pruning to improve accuracy, it can automatically handle unbalanced data samples, and has strong fit and stability. and got results.

Gradient Boosted Trees

The figure above shows the workflow of a gradient boosting tree classifier. This method reduces the predicted value, and by repeatedly changing the sample difference, the trained model has different decision-making capabilities. When training a model, the smaller the loss of prediction samples, the better, and the underlying model must be a decision tree. and gives the result.

5. The best classifier that you selected – the type, its performance, how it solved the problem (if it makes sense for that type of classifier), and reasons for selecting it.

Among these methods, I think the random forest classifier is the best of the six, so I optimized some of them. To get more accurate predictions, I performed round-robin operations in Random Forest Learner and Random Forest Predictor. For the UnknownData.csv filcreateI use a new file read node to import. Next, I use the same Number to String app node and Normalizer app node to make sure the same preprocessing method is used. I connected the normalize application node to the random forest predictor node and in the random forest predictor changed the predicted column name to Prediction (AIRLINE_NAME), used the column filter to select Row ID and Prediction (AIRLINE_NAME), then in the CSV Name the filename. Write and export files.