Preparing The Data:
* First of all, we loaded the Iris dataset from the scikit-learn library.
* We printed the dataset, feature names & labels (target names).
* We split the data (features, labels) into 70% training, 30% testing.
Decision Tree Algorithm:
* We defined a decision tree classifier.
* We trained the model using 'fit' function on the 70% data we put aside for training.
* We tested the X_test data we put aside for testing, using 'predict' function & got predict_dt.
* We checked the accuracy of our results against the y_test data we put aside - multiplied by 100 to get %,
and got accuracy of approx. 95%.
* We visualized the decision tree.
KNN Algorithm:
* We scaled down the 70% feature data (X_train, X_test), so we'll have all data in range.
That is needed in KNN algorithm, because it's training relays on minimum distance from specific point in range.
* We defined a KNN classifier.
* We trained the model using 'fit' function on the 70% scaled down features data and 70% labels data for training.
* We gave a default n_neighbors = 5 that gives higher accuracy results.
* We tested the X_test scaled down data using 'predict' function & got predict_knn.
* We checked the accuracy of our results against the y_test data we put aside - multiplied by 100 to get %,
and got accuracy of approx. 97%.
Conclusion
* Classification - a type of supervised ML problem where the target variable is categorical.
Given a training data that contains a known label, the classifier maps from the input to output variables.
* There are a number of ways to classify the Iris Flower data set, but I chose to learn Decision Tree and KNN
that are both supervised ML algorithms.
* Decision Tree - builds a classification model on the training dataset before being able to actually classify
an observation from test dataset. Then the model is able to classify unseen training instances (observations).
* KNN - directly learns from the observations. It starts processing data after it is given
a test observation to classify. It is not recommended to use when there's a large number of features.
* There are other supervised algorithms we could have used for classification, such as:
Naive Bayes, LDA, QDA, SVG.