Iris - Most loved dataset

Iris dataset

1. Iris Dr Asmita Titre

2. About Iris Dataset Best known database to be found in the pattern recognition literature. Data set- Iris flower data set(Donated date - 1988-07-01), also known as Fisher's Iris data set and Anderson's Iris data set b/c Edgar Anderson collected the data. It is multivariate(more than 2 dependent variable) data set Study of three related Iris flowers species. Data set contain 50 sample of each species(Iris- Setosa , Iris- Virginica , Iris- Versicolor )

3. Features or attributes (4 numeric)- Sepal length in cm Sepal width in cm Petal length in cm Petal width in cm

4. More on data set- One class is linearly separable from the other 2; the latter are NOT linearly separable from each other Missing Attributes Values : None

5. Summary Statistics Class Distribution: 33.3% for each of 3 classes.

6. Iris Virginica , Iris Versicolor , Iris setosa

7. Objective Classify a new flower as belonging to one of the 3 classes given the 4 features

8. Lets Be Sherlock Holmes 

9. What is data saying ? ( Exploratory data analysis). We will try to find the answer of the following questions with the help of all available asset

10. 1. Descriptive statistics- SD, Min, Max etc 2 . Class Distribution (Species counts are balanced or imbalanced) – Balanced 3 . Univariate Plots:- Understand each attribute better.

11. 3.1 Box Plot - Distribution of attribute through their quartiles & find outlier # Box and whisker plots(Give idea about distribution of input attributes)

12. 3.2 Histogram -

13. 3.2 Histogram 3.2 Distribution of attribute through their bin, we find the distribution of attribute follow Gaussian or other distributions

14. 4. Multivariate Plots Understand the relationships between attributes & species better. (Which attributes contributes a lot in classifying species)

15. 4.0 Scatter Plot - Sepal_Length_Width Vs Species.

16. Observations- 1.Using Sepal_Lenght & Sepal_Width features, we can only distinguish Setosa flower from others 2.Seperating Versicolor & Virginica is much harder as they have considerable overlap 3.Hence, Sepal_Lenght & Sepal_Width features only work well for Setosa

17. 4.1 Scatter Plot - Petal_Length_Width Vs Species.

18. Observations- 1.Using Petal_Lenght & Petal_Width features, we can distinguish Setosa , Versicolor & Virginica fairly 2.There are slightly overlap of Versicolor & Virginica . 3.Graph shows that Petal (Length and Width) features are best contributor for Iris Species as compare to Sepal (Length and Width)

19. 4.2 Scatter Plot of all the attributes

20. 4.3 3D Plot

21. 4.4 Violinplot - Density of the length and width in the species

22. Machine Learning (Here comes the beauty of machine)

23. Steps to implement the ML 1 Import Library 2 Create Correlation Matrix 3 Spliting the Data Set But keep in mind;

24. While Spliting the Data Set 3.1 Take all the data features 3.2 Take only Sepal Features(Length & Width) 3.3 Take only Petal Features(Length & Width) 3.4 Take all relevant Features from correlation Matrix

25. Algorithms used 4 Evaluate by using 6 different Algorithms(Cross Validation) Here, 4.1 Logistic Regression (LR) 4.2 Linear Discriminant Analysis(LDA) 4.3 K-Nearest Neighbour (KNN) 4.4 Classification and Regression Tree(CART) 4.5 Gaussion Naive Bayes (NB) 4.6 Support Vector Machine

26. 5 Final Evalution (Compare all model according to features selection and accuracy) 6 Deep Learning

27. Final Evaluation Note:- Testing dataset size(validation size) is small

28. Final Evaluation -cont Case Features used Best Model Train Accuracy Test Accuracy Missclassified 1 All features in SVM .9899 .9555 2 classes 2 Sepal only SVM .8472 .7111 12 3 Petal only SVM .9899 .9333 3 4 PetalWidth,Sepal ( Len,Wid ) SVM/LDA .9809 .9111 4 5 PetalLen,Sepal ( Len,Wid ) SVM .9700 .9111 4

29. Thank you…