Machine learning (ML) was defined as “the field of study that gives computers the ability to learn without being explicitly programmed” by Arthur Samuel. It was evolved from pattern recognition and computational learning in artificial intelligence. ML mainly focuses on developing models which are capable of independently adapting to changes when exposed to new data.
The requirements to create a good Machine Learning system includes data preprocessing capabilities, Choosing the correct algorithms, automation and iterative processes, scalability and ensemble modelling to improve the performance of the models.
ML tasks are classified into three based on the nature of learning. They are,
- Supervised learning
- Unsupervised learning
- Reinforcement learning
Supervised learning algorithms uses the given set of predictors (independent variables) to predict the target variable (dependent variable). The algorithm learns by comparing the actual output with the predicted outputs to find the errors, this process is called Training. The model is modified according to the trained data. Regression and classification are two major subcategories of supervised ML. Supervised learning is used in applications which require predicting future events based on historical data.
Unsupervised learning does not have target variable (dependent variable) to predict. They are used for exploring the data to find out some relationship within the data. There are no training data used in unsupervised learning. The system uses the given data to find out correlations and discovers patterns within the data.
Reinforcement learning algorithms makes use of trail and error method to find out which actions yield the maximum output. The machine learns from its past experience and applies the learning in making accurate decisions. These algorithms are used for gaming, robotics and navigation.