Getting Started with Machine Learning !

Original article was published on Artificial Intelligence on Medium

Ever wondered what machine learning is, metaphorically it means to make a machine to learn, predict and provide insights like human.

In recent technology machine learning is nothing new. There are three different kinds of learning to deal with complex data that trains a machine to work like a human. Obviously with advantages and disadvantages that is the cost of everything.

1. Supervised Learning

Like a class of students and a professor, here the students are trained by professors.

In supervised learning the data is labelled or annotated for example the input are images of apple and an annotation is given that these are apple so that the model predicts “This is an Apple” .

Supervised Learning Model

Supervised learning further can be classified into two different categories that is shown below.

Types of supervised learning


It is a task to predict continuous numeric quantity and utilized to answer the factors which are most correlated, helps in deciding what to do take it or not and find the root cause of why this happened. Remember “Don’t mistake correlation for causation, its not” . We use root mean squared error better known as RMSE value to evaluate the predictions.


It is a task to predict a discrete target label (also known as class label) you want to predict based on other existing attributes. Rather it predicts labels or categories for the new data for example using it for email spam filter “spam or not”, predict sentiment “happy or not happy” or predict the payment behavior. Accuracy term is used to evaluate the model.

The advantages and disadvantages are described below for supervised learning.


1. Accuracy is really high as the input data is labelled data.

2. The modelling is very specific as the data is annotated.

3. Supervised model are very useful in classification.


1. Its a time-consuming process as to train data the computational time is really high.

2. It doesn’t happen in real-time.

3. The method gets complex in each steps performed.

Now we are going to talk about the rest of the machine learning models.

2. Unsupervised Learning

Like a class full of students asked to learn without a professor, this is unsupervised learning.

In unsupervised learning the input data is not at all labelled as apple or banana, the model recognizes the pattern and categorize it accordingly as shown below.

Unsupervised Learning Model

Unsupervised learning further can be classified majorly in two different categories that is shown below.

Types of unsupervised learning


It is a category of objects that falls into similar class, known as cluster. These group of cluster known as clustering. There are two different methods by which clustering is carried by is hierarchical and partition clustering. Its used for outlier detection for fraud, image processing, recommender system or market research and many other applications.

Dimension Reduction:

It is usually practiced on highly dimensional data to reduce its dimension as the name suggests. It usually reduces the number of features that are considered in the data as its directly effects the sparsity of data that should be low. There are two methods that are widely used are feature selection and feature extraction. Its used widely for intrusion detection, face recognition, customer relation management, protein analysis and many more.

Now moving onto advantages and disadvantages of unsupervised learning.


1. It has potential to unravel any hidden pattern and recognize it.

2. Very efficient for running over large and complex data.

3. It doesn’t lose its effectiveness on accuracy over missing data or on large missing data.

4. It reduces the possibility of human error as the data is not labelled.


1. There is no precise information about data sorting and the output of the data as its unlabeled.

2. It consumes more time while interpreting the data.

3. Overfitting is a major problem in unsupervised learning.

4. The spectral class doesn’t always correspond to informational classes.

3. Reinforcement Learning

This is used to drive through decision making process. The output depends upon the previous or current input and the next step is dependent on the output obtained recently. It can be said it needs feedback of previous output at every step to decide. The best application of reinforcement learning is mobile games or computer game like cross-zero, chess any game that is against a non-human. It works on two different methodology that is positive and negative.


1. Used to resolve long term and complex problem that can not be achieved by classical techniques.

2. It can correct the errors caused during the learning process as it works on mechanism similar to feedback one.

3. Chances of reoccurring error in the model is negligible.


1. It requires a lot of computation and lots of data.

2. The results can also be diminished if it gets into overload state due to much of reinforcement.

3. Not at all feasible for solving less complex problem.