Question-66: Can you graphically show the underfitting model?

Answer: As you can see in below image the 3^rd graph is not able to correctly predict the data with even the training dataset. Hence, it is considered underfitting. (check HadoopExam.com machine learning training to understand further)

Question-67: Why model overfitting happens?

Answer: Because your model tries to memorize more and more data during the training data set.

Question-68: If our model is underfitting and poorly perform what we can do?

Answer: When your model perform badly with the training dataset then possible reason could be your model is very simple and your input data features are not expressive enough to describe the target output. And to improve the performance in this case can be done by increasing the flexibility of the model and to increase the model flexibility we can take following steps

• Add new features in the input training dataset.
• Reduce the regularization in the data.

Question-69: What should you do reduce the model overfitting?

Answer: To avoid overfitting with the training dataset, you can reduce the model flexibility. And to reduce the model flexibility, we can take following actions.

• Reduce the number of feature combinations.
• Reduce the number of numerical attribute bins.
• Increase the amount of regularization is used.

Question-70: What other options do you see to increase the accuracy of the model?

Answer: To increase the accuracy of the model you should not have poor data and to improve it, you can do the following things.

• Increase the amount of training data examples.
• Increase the number of iterations on the training data.

Question-71: What is Machine Learning, can you explain once again?

Answer: Using Machine Learning on the available historical data, you or your team can make better business decisions. You use different Machine Learning algorithms for different needs to find the patterns in the available data. These models are mathematical. Once the correct or near accurate model you find then you put this model in productions to make the predictions on the new or future data.

Question-72: Can you give an example, when you don’t want to use Machine Learning for predicting the future values?

Answer: We should not use the Machine Learning when you can determine the future value based on some simple or existing solutions like applying some rules using rule engine or by simply doing calculations you can derive the value or you can write step by step program to derive the value.

Question-73: Can you give the example when you really want to use Machine Learning?

Answer: You don’t have choice and need to go for Machine Learning only, in following situations

• When you are not able to use rules-based algorithm to derive the outcome on new data.
• When your computation can not derive the target value.
• When your computation is depending on too many factors and you can not code them in program.
• When you have to predict outcome on large volume of data. Like predicting website visitor will buy a product or not when there are millions of visitors on the website.