Machine learning is an application of Artificial Intelligence (AI). Then what is AI? Artificial intelligence is the process of making machines capable of performing human tasks. As we humans learn from our past experiences, machines learn from past data. We tend to respond to a situation based on our past experiences. In our brain, we have stored all the situations we had encountered and the corresponding responses as a key-value pair (A key-value pair (KVP) is a set of two linked data items).
Our brain can recognize a pattern in each of these situations and responses. So whenever we encounter a new situation, the brain looks up the stored key-value pair list and pulls out the response corresponding to the pattern closest to the new situation.
Likewise, machines analyze and identify patterns in historical data. Based on the patterns machines can predict the responses for a new set of data. Thus machine learning (ML) is the process of learning from data and making predictions useful for future decision-making.
Email spam detection, traffic modeling, speech recognition, climate prediction, and so on, to name a few, are the various areas where we can apply ML algorithms. There are different algorithms in ML which analyze and learn from historical data in an iterative manner and predict future behavior.
ML can be used to solve a problem if, in general, the following 3 scenarios are present:
It cannot be described mathematically
There exists a pattern
There is data
Even so, the first two conditions are not essential. We can still use ML techniques if we prefer, for problems that are mathematically definable. When no pattern exists it may be difficult, but these techniques can still be used. However, the last point is the most important: it is imperative that data is available; without data, we cannot do any analysis.
After identifying the problem, we need to collect and prepare the data before we feed it to the machines.
After preparing the data we split the data into two parts: training data and test data. An appropriate algorithm is identified to build the model with the training data, and then we evaluate the model using the test data.
After evaluating the model the performance of the model can be improved, if required. This is done by either selecting a new algorithm or by adjusting the training data.
The three main types of algorithms in ML, based on available data and the problems in each of the types are:
As the names suggest, Supervised Learning involves learning with some supervision from an external source whereas Unsupervised Learning does not. Reinforcement learning( RL)is learning by interacting with an environment. An RL agent learns from the consequences of its actions. It selects its actions on basis of its past experiences (exploitation) and also by new choices (exploration), which is essentially trial and error learning.
Hope you got an overview of ML. I would like to share with you a few tips which would help you to learn machine learning within a short span of time.
Start with algorithms in Supervised Learning. I believe it is the simplest one to understand. i.e. Simple Linear Regression and Multiple Linear Regression for regression analysis. Logistic regression and K-nearest neighbors for classification. Then move to Unsupervised Learning. i.e.k-means and hierarchical clustering for clustering tasks.
Once you understand the goals of each of the above algorithms then it is easy to move forward and explore more algorithms.