Scribd
Upload
151 views

Cross-Validation in Machine Learning

Cross-validation is a technique used to evaluate machine learning models on a limited data sample. It involves splitting the data into training and validation sets, training the model on the training set and evaluating it on the validation set. Common cross-validation methods described in the document include hold out validation, k-fold cross-validation, leave one out cross-validation, and stratified k-fold cross-validation. These methods are implemented in Python using scikit-learn to effectively test machine learning models.

Uploaded by

Priya dharshini.G
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
151 views

Cross-Validation in Machine Learning

Cross-validation is a technique used to evaluate machine learning models on a limited data sample. It involves splitting the data into training and validation sets, training the model on the training set and evaluating it on the validation set. Common cross-validation methods described in the document include hold out validation, k-fold cross-validation, leave one out cross-validation, and stratified k-fold cross-validation. These methods are implemented in Python using scikit-learn to effectively test machine learning models.

Uploaded by

Priya dharshini.G
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 18

Cross-Validation in Machine

Learning
• Cross-validation is a technique for validating the model efficiency by
training it on the subset of input data and testing on previously unseen
subset of the input data. We can also say that it is a technique to check
how a statistical model generalizes to an independent dataset.
• In machine learning there is always the need to test the stability of the
model. It means based only on the training dataset; we can't fit our
model on the training dataset. For this purpose, we reserve a particular
sample of the dataset, which was not part of the training dataset. After
that, we test our model on that sample before deployment, and this
complete process comes under cross-validation. This is something
different from the general train-test split.
• Hence the basic steps of cross-validations are:
• Reserve a subset of the dataset as a validation set.
• Provide the training to the model using the training dataset.
• Now, evaluate model performance using the validation set. If the model
performs well with the validation set, perform the further step, else
check for the issues.
Key aspects of evaluating the quality of the model are –
• How accurate the model is
• How generalized the model is
• When we start building a model and train it with the ‘entire’ dataset, we can very well calculate its accuracy
on this training data set. But we cannot test how this model will behave with new data which is not present in
the training set, hence its generalization cannot be determined.
• Hence we need techniques to make use of the same data set for both training and testing of the models.
• In machine learning, Cross-Validation is the technique to evaluate how well the model has generalized and its
overall accuracy. For this purpose, it randomly samples data from the dataset to create training and testing
sets. There are multiple cross-validation approaches as follows –

• 1.Hold Out Approach


• 2.Leave One Out Cross-Validation
• 3.K-Fold Cross-Validation
• 4.Stratified K-Fold Cross-Validation
• 5.Repeated Random Train Test Split
• 1. Hold Out Approach
• In the hold-out approach, the data set is split into the train and test set with random sampling.
The train set is used for training the model and the test set is used to test its accuracy with
unseen data. If the training and accuracy are almost the same then the model is said to have
generalized well. It is common to use 80% of data for training and the remaining 20% for testing.
• Advantages
• It is simple and easy to implement
• The execution time is less.
• Disadvantages
• If the dataset itself is small, setting aside portions for testing would reduce the robustness of
the model. This is because the training sample may not be representative of the entire dataset.
• The evaluation metrics may vary due to the randomness of the split between the train and test
set.
• Although 80-20 split for train test is widely followed, there is no thumb rule for the split and
hence the results can vary based on how the train test split is done.
• 2. Leave One Out Cross Validation (LOOCV)
• In this technique, if there are n observations in the dataset, only one observation is
reserved for testing, and the remaining data points are used for training. This is
repeated n times till all data points have been used for testing purposes in each
iteration. Finally, the average accuracy is calculated by combining the accuracies of
each iteration.
• Advantage
• Since every data participates both for training and testing, the overall accuracy is
more reliable.
• It is very useful when the dataset is small.
• Disadvantage
• LOOCV is not practical to use when the number of data observations n is huge. E.g.
imagine a dataset with 500,000 records, then 500,000 model needs to be created
which is not really feasible.
• There is a huge computational and time cost associated with the LOOCV approach.
• 3. K-Fold Cross-Validation
• In the K-Fold Cross-Validation approach, the dataset is split into K folds. Now in 1st iteration, the
first fold is reserved for testing and the model is trained on the data of the remaining k-1 folds.
• In the next iteration, the second fold is reserved for testing and the remaining folds are used for
training. This is continued till the K-th iteration. The accuracy obtained in each iteration is used
to derive the overall average accuracy for the model.
• Advantages
• K-Fold cross-validation is useful when the dataset is small and splitting it is not possible to split
it in train-test set (hold out approach) without losing useful data for training.
• It helps to create a robust model with low variance and low bias as it is trained on all data
• Disadvantages
• The major disadvantage of K-Fold Cross Validation is that the training needs to be done K times
and hence it consumes more time and resources,
• Not recommended to be used with sequential time series data.
• When the dataset is imbalanced, K-fold cross-validation may not give good results. This is
because some folds may have just a few or no records for the minority class.
• 4. Stratified K-Fold Cross-Validation
• Stratified K-fold cross-validation is useful when the data is imbalanced.
While sampling data into K-folds it makes sure that the distribution of all
classes in each fold is maintained. For example, if in the dataset 98% of
data belongs to class B and 2% to class A, the stratified sampling will
make sure each fold contains the two classes in the same ratio of 98% to
2%.
• Advantage
• Stratified K-fold cross-validation is recommended when the dataset is
imbalanced.
• 5. Repeated Random Test-Train Split
• Repeated random test-train split is a hybrid of traditional train-test
splitting and the k-fold cross-validation method. In this technique, we
create random splits of the data into the training-test set and then
repeat this process multiple times, just like the cross-validation
method.
Examples of Cross-Validation in Sklearn Library
• About Dataset
• We will be using Parkinson’s disease dataset for all examples of cross-validation in the Sklearn library. The goal is to predict
whether or not a particular patient has Parkinson’s disease. We will be using the decision tree algorithm in all the examples.
• The dataset has 21 attributes and 195 rows. The various fields of the Parkinson’s Disease dataset are as follows –
• MDVP:Fo(Hz) – Average vocal fundamental frequency
• MDVP:Fhi(Hz) – Maximum vocal fundamental frequency
• MDVP:Flo(Hz) – Minimum vocal fundamental frequency
• MDVP:Jitter(%),MDVP:Jitter(Abs),MDVP:RAP,MDVP:PPQ,Jitter:DDP – Several
• measures of variation in fundamental frequency
• MDVP:Shimmer,MDVP:Shimmer(dB),Shimmer:APQ3,Shimmer:APQ5,MDVP:APQ,Shimmer:DDA – Several measures of
variation in amplitude
• NHR,HNR – Two measures of ratio of noise to tonal components in the voice
• status – Health status of the subject (one) – Parkinson’s, (zero) – healthy
• RPDE,D2 – Two nonlinear dynamical complexity measures
• DFA – Signal fractal scaling exponent
• spread1,spread2PPE – Three nonlinear measures of fundamental frequency variation
• Importing Necessary Libraries
• We first load the libraries required to build our model.

• import pandas as pd
• import numpy as np
• from sklearn.tree import DecisionTreeClassifier
• from sklearn.model_selection import train_test_split
• from sklearn.model_selection import KFold
• from sklearn.model_selection import StratifiedKFold
• Reading CSV Data into Pandas
• Next, we load the dataset in the CSV file into the pandas dataframes
and check the top 5 rows.
• df=pd.read_csv(“Parkinsson disease.csv")
• df.head()
• Data Preprocessing
• The “name” column is not going to add any value in training the model
and can be discarded, so we are dropping it below.
• df.drop(df.columns[0], axis = 1, inplace = True)
• Next, we will separate the feature and target matrix as shown below.
• #Independent And dependent features
• X=df.drop('status', axis=1)
• y=df['status']
Hold out Approach in Sklearn

• The hold-out approach can be applied by using train_test_split module of


sklearn.model_selection
• In the below example we have split the dataset to create the test data with a size of
30% and train data with a size of 70%. The random_state number ensures the split is
deterministic in every run.
• from sklearn.model_selection import train_test_split
• X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.30, random_state=4)
• model = DecisionTreeClassifier()
• model.fit(X_train, y_train)
• result = model.score(X_test, y_test)print(result)
• Out[38]:
• 0.7796610169491526
K-Fold Cross-Validation
• K-Fold Cross-Validation in Sklearn can be applied by using cross_val_score module
of sklearn.model_selection.
• In the below example, 10 folds are used that produced 10 accuracy scores using which we
calculated the mean score.
• In [40]:
• from sklearn.model_selection import cross_val_score
• model=DecisionTreeClassifier()
• kfold_validation=KFold(10)
• results=cross_val_score(model,X,y,cv=kfold_validation)
• print(results)print(np.mean(results))
• Out[40]:
• [0.7 0.8 0.8 0.8 0.8 0.78947368
• 0.84210526 1. 0.68421053 0.36842105]
• 0.758421052631579
• Stratified K-fold Cross-Validation
• In Sklearn stratified K-fold cross-validation can be applied by
using StratifiedKFold module of sklearn.model_selection
• In the below example, the dataset is divided into 5 splits or folds. It returns 5
accuracy scores using which we calculate the final mean score.
• from sklearn.model_selection import StratifiedKFold
• skfold=StratifiedKFold(n_splits=5)
• model=DecisionTreeClassifier()scores=cross_val_score(model,X,y,cv=skfold)
• print(scores)print(np.mean(scores))

• Out[41]:
• array([0.61538462, 0.79487179, 0.71794872, 0.74358974, 0.71794872])
• 0.717948717948718
Leave One Out Cross Validation(LOOCV)

• In Sklearn Leave One Out Cross Validation (LOOCV) can be applied by using LeaveOneOut module of sklearn.model_selection
• from sklearn.model_selection import LeaveOneOut
• model=DecisionTreeClassifier()
• leave_validation=LeaveOneOut()
• results=cross_val_score(model,X,y,cv=leave_validation)
• results

• Out[22]:
• array([1., 1., 1., 1., 1., 0., 1., 1., 1., 1., 1., 1., 1., 0., 1., 1., 1.,
• 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 0., 1., 1., 1., 1.,
• 1., 1., 1., 1., 1., 0., 0., 1., 1., 1., 0., 1., 1., 1., 1., 1., 1.,
• 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
• 0., 1., 1., 1., 1., 0., 1., 1., 1., 1., 1., 1., 1., 0., 1., 1., 1.,,,,,,,,,,,,,,,,,,,,,,,,,,,]
• print(np.mean(results))
• Out[44]:
• 0.8358974358974359
Repeated Random Test-Train Splits

• In Sklearn repeated random test-train splits can be applied by using ShuffleSplit module


of sklearn.model_selection
• In [45]:
• from sklearn.model_selection import ShuffleSplit
• model=DecisionTreeClassifier()
• ssplit=ShuffleSplit(n_splits=10,test_size=0.30)
• results=cross_val_score(model,X,y,cv=ssplit)
• print(results)print(np.mean(results))
• Out[45]:
• array([0.79661017, 0.71186441, 0.79661017, 0.88135593, 0.72881356,
• 0.84745763, 0.83050847, 0.77966102, 0.83050847, 0.81355932])

You might also like