Question

Implement the following machine learning tasks, utilizing Regression techniques e Prediction e Classification eFeature Reduction Feature Independence Model Selection (underfitting and overfitting analysis) 2 Required Components You may utilize Python or Matlab libraries, for the following implementations. 1. Implement prediction utilizing multiple linear regression on a data set with several features Perform an evaluation of the residuals to check for assumptions of your model, such as li earity, noise term with zero mean and constant variance, normality and so forth. 2. Implement feature reduction, utilizig hypothesis testing and p-values on the standard er- ror, for each estimated parameter. Therefore, if an estimated parameter is not statistically different from zero, the associated feature maybe excluded from the model. 3. Test your algorithms and implementations on two data sets from the UCI, Kaggle or the ImageNet repository. One of these data sets can have less than 5 features. The second data set should have at least 10 features. (a) Some example data sets for the prediction task are: Boston Housing Data to predict the median price of housing Wine Quality Data to predict the quality of wine Million Song data to predict the release year (select a sample) (b) Some example data sets for the classification task are: . Classify email as spam or not spam MNIST handwritten digit classification (you can select 2 out of the 10 digits) Image object classification. Medical diagnostics.

Answer 1

The Machine Learning Task which can be performed using classification is Credit Card Fraud Detectionwhere we will classify whether a particular transaction by the customer is a fraud transaction or not on the basis of the previous transactions which are done by the Customer.

In this task the inputs will provided by a particular dataset which is available very easily online at various platforms in the form of csv files.(e.g. creditcard.csv). Once we have the data, we can specifically tell about the inputs or the labels but some which are present are Time of the transactions, Valid Transactions, Fraud Cases etc.

The output of this task will be first to analyze the dataset using our Machine Learning algorithm and thenpredicting whether a given transaction from the test set is valid or fraud based upon the training set and the training which has taken place.

The goal will be to design our algorithm in such a way that we get high prediction accuracy or the classification accuracy of test dataset as valid or fraud transactions i.e. if a transaction is a fraud transaction it should be classified as fraud and valid transaction if it is actually a valid one.

For data preparation, we can do the data cleaning which involves filling up the missing values, removing noisy data, smoothing out data types inconsistencies and removing outliers.We can also go for data reduction in which we can remove data which is insignificant with respect to our result obtaining process.Data transformation is another technique which can help in data preparation, through this we cantransform large values into smaller values or integers into float for obtaining precise and quick result.

For collecting the training data, what we can do is that dataset which we will use can be divided into two parts i.e. in the ratio of 70:30 where 70% of the dataset will be training set and 30% will be test set. The ratio can vary a bit and even we can try different proportions to obtain better accuracy. But it is recommended to use 70-80% of the dataset as training set otherwise we may suffer from the problem of underfitting or overfitting of the data.

The ground-truth label for all samples is obtained through the training which we will be providing to our machine learning model, in this our model will understand the various kinds of labels which are required in the output of the samples which are provided.