What is Machine Learning ?

Machine learning gives computers the ability to learn and perform certain tasks without being explicitly programmed. That means, your computer will learn from data itself , you don’t need to write a program again and again for different datasets. Write once, use many times.Once you develop a ML model like Linear Regression, Decision Tree, Random Forest, Logistic Regression, so on. The only thing you need to train it again and again if your Model performance is degrading or you can also automate or schedule your model to get training time to time on a regular basis. It all depends upon you as a Data scientist, ML engineer or Data Analyst , whatever you call yourself.

Machine Learning is one of the most in demand skills of the 21st century. It has a number of applications that we use in day to day life such as YouTube video recommendations, Facebook friends suggestion, Netflix Movie recommendations, etc.

Machine Learning was first coined by Arthur Samuel in 1959. He was an American pioneer. He defined ML as “a field of study that gives computers the ability to learn without being explicitly programmed”. 

There are multiple different ways to understand Machine learning : 

• ML is a computer program that has the ability to learn and store patterns from the past or Historical Data and can predict if we give input so and so. The inputs we insert into the Model at the time of training must be the same at the time predicting something.
• ML can learn from the experience with respect to some class labels.

For example, let’s say you want to predict whether a customer should be given a loan or not, here the class label is Yes or No (Yes – should provide loan , No – shouldn’t provide loan). To predict Yes or No , we need some input Data to train my model. Input Data can be customer’s Income, credit score, Age, Profession, Education Background, family background, demography, etc. Once you collect these data, you can train your ML Model. After get trained, when you will ask the model, by inserting the input data like  income is 50,000 $ ,credit score is 400, Age is 35, Education is Computer science , and so on, the Model will respond in Yes or No, and this depends upon its learning.If the Model response is in Yes, the person is eligible for getting a loan else Not.

• Machine learning is a subset of Artificial Intelligence which provides machines the ability to learn automatically and improve from experience without being explicitly programmed. It is a practice of getting machines to solve complex problems by gaining the ability to think. This ability gained by machines from past data itself just like humans has ability to think and make decisions based on the past from the subconscious  Mind. In general, ML models also create a kind of subconscious mind to store its learning so that it could predict / respond when you ask something.
• Machine learning algorithm is a set of rules and statistical techniques used to learn patterns from the data and draw significant information from them. So, an algorithm is the logic behind the Machine learning algorithm.
• When we insert input data or input Data with its label or class into a Machine learning algorithm while training, the algorithm creates a Model that we can save as an object file. Model is the main component of the algorithm. It is the model that we train data using the Machine learning algorithm and the algorithm is going to map all the decisions that the model is supposed to take based on the given input in order to get the correct output. Input data is independent variables whereas output data is a dependent variable

For example, sanction of loan is my output or dependent variable which depends on some factors like your income, education, credit score, profession, etc,  these factors are nothing but input or independent variables.

Now understand Machine Learning processes step by step: 

• Machine learning involves building a model that can be used to find the solutions for any problem statement. The problem statement is the very first task that is given to you to build a Model. 
• To solve the problem you need data –

if you have data then that’s fine, if not, you have to collect it. Data you can collect from multiple sources depending upon what kind of problems are going to solve whether it is just a kaggle problem or its an industry’s complex problems.

If you want to practice your Machine learning skills,you will get data easily. There are multiple ways to do this , kaggle is one of the examples. But if you are going to solve an industry’s problems, data is given by your client or if your company itself is a source of data generation, then data is in your hand. 

Sometimes you need to buy data for your experiment and research. You can also collect data from multiple websites called web scraping. 

• After collecting Data –

Once you collect data, data in its raw form is not always rich in information. It has different kinds of errors like data redundancy, Missing value, outlier, data invalid. You have to fix this by applying different techniques.

• After data cleaning –

you need to explore the data to find out the trends or insights by different visualization techniques such as line charts, Bar graphs, Histogram, Pie chart , etc.

• Before training the model- 

you also need to do Feature engineering, feature selection, feature scaling.

• After feature selection –

you are ready to build the model. Model Building is not a big deal. You need a few lines of code if you are using R or Python.

• Now model evaluation required –

you need to check your model’s performance on the test Data using Confusion Matrix, RMSE, MSE, R2, adjusted R2 depending upon your kind of problems.

• Model deployment – 

deployment Once your Model accuracy is up to the mark, you are ready to deploy in production. 

• That is how you create any machine learning model.

There are four types of Machine learning : 

1. Supervised learning
2. Unsupervised learning
3. Semi Supervised learning
4. Reinforcement learning

1.Supervised learning : 

Supervised learning as the name indicates the presence of a supervisor as a teacher. Basically supervised learning is a learning technique in which we teach or train the machine using data(set of training data) which is well labeled that means some data is already tagged with the correct answer. This labeled data is called a dependent variable. After that, the Model is provided with a new set of examples(set of test data). Model returns an output which is a predicted value. Now we can test its performance by comparing the predicted values with the actual. The actual value is present in the set of test data. The performance is determined by calculating RMSE, R2, or Confusion Matrix, Precision, Accuracy, etc.

2.Unsupervised learning : 

Unsupervised learning is the training of machines using information that is neither classified nor labeled with any class. Unlike supervised learning, It learns without any guidance and training because there is no teacher or no supervisor (no output /dependent data) involved here. It sorts the unsorted data into some groups with similar characteristics. The group is also called cluster. The number of clusters you can choose while inserting input data into the model. The algorithm we use here is k-means clustering, hierarchical clustering, etc. If You use K-means, you can use the Elbow method for selecting the right clusters.  Dendrogram is suitable for hierarchical clustering.

3.Semi supervised Learning: 

Semi supervised is a combination of Supervised and unsupervised learning techniques. This is generally used if there is a lack of labeled data in the dataset. To solve this kind of problem we use unsupervised techniques to make clusters of data and generate the label of each row of the dataset. This way we can create a label of each row and  can train like a supervisor.

The examples are Internet content classification and Speech analysis.

4.Reinforcement Learning: 

Reinforcement Learning has a little different approach then supervised or unsupervised. It takes a suitable action to maximize reward in a particular situation. Whenever it performs any task, it expects the rewards or punishment in order to test itself whether it has performed well or not. If on every task machines get rewarded that means it performs well otherwise it needs to be improved. Therefore, Reinforcement learning is bound to learn from its experience and improves our-self because there is no training given like supervised learning. 

 Thank You and Best of luck