Machine learning is commonly defined as a field of computer science that gives machines the ability to learn without being explicitly programmed.

The above statement although correct may not provide clear explanation to someone new to this field.

Lets first understand what we mean by the term ‘ability to learn’. In the context of machine learning, it can be considered as a process of applying algorithms over a dataset in order to find patterns which is then used to predict the outcome on a new dataset. This is also called as model training.

The outcome of the training process is a model. In layman terms consider model as a function which when provided some data as input will give you back a response. The data provided has to be in the same form and format as the one used in the model training.

Types of Machine learning algorithms:

1. Supervised: The algorithm is trained on a labeled data i.e. data with known output. For e.g. cancer detection where output could be benign or malignant. The problems solved under supervised learning can be categorized as:
   • Classification problems where the labels in the dataset are discrete. For example, Emails can be classified as Spam or Not Spam. The problem could be binary or even multiclass i.e. more than two output values are expected
   • Regression problems where the labels in the dataset are continuous. For e.g., predicting price of any product is a regression problem
2. Unsupervised: The algorithm have to work with dataset without labels. The objective of these algorithms is finding different kinds of relation. Look at the following types of unsupervised problems which can clarify what kind of problems are dealt with these algorithms:
   • Clustering problems where datasets are categorized into ‘n’ number of groups. For e.g. customer segmentation problem for targeted marketing is a clustering problem
   • Recommendation problem where association is derived between various product types so that customers can be presented with upsell or cross-sell products
3. Reinforcement: These algorithms are about training a model to make a sequence of decisions where at each step the decision taken is either rewarded or penalized depending on the favorability of the outcome.

If you find this topic interesting, you can find more blogs on it here Machine Learning

.