How does machine learning works?

Machine learning is the part of Artificial Intelligence, which teaches computers to think about ways of thinking similar to humans, such as learning and improving from past experiences. It works by exploring, investigating data and identifying patterns and involves minimal human intervention. A key aspect of what makes machine learning valuable is it's ability to discover what and which information has been left out of human eyes while collecting data. Machine learning helps in determining the complex patterns that can potentially be ignored in human analysis. To understand how machine learning works, it is very important to understand it's algorithms.

1. Supervised machine learning
2. Unsupervised machine learning
3. Semi-supervised machine learning
4. Reinforcement machine learning

1. Supervised machine learning

   In this algorithm, the machine reads from the past experiences, and applies it to the new data. So that it can predict the events that will happen in the future by using the previous given examples. This algorithm works in exactly the same way that humans actually learn with their past experiences. In supervised learning, various types of examples and answers are given as input to the machine; this algorithm learns from these examples and guesses the correct output based on the given inputs.

   Supervised machine learning is further classified into:

   1. Classification

      The machine learning must draw conclusion from existing data and should predict to which class new data belongs. For example, new emails are filtered as spam and non-spam based on previous data.

      Most popular classification algorithms are:

      • Naive Bayes
      • Logistic Regression
      • KNN Model
      • Support Vector Machines (SVM)

   2. Regression

      In this algorithm, the machine is taught to identify and predict relationship among one dependent and other series of independent variables.

      Some regression algorithms are:

      • Linear Regression
      • Ridge Regression
      • Neural Network Regression
      • Lasso Regression
      • Decision Tree Regression
      • Random Forest
      • KNN Model
      • Support Vector Machines (SVM)
      • Dimensionality Reduction Algorithms
      • Gradient Boosting algorithms
        1. GBM
        2. XGBoost
        3. LightGBM
        4. CatBoost

   3. Forecasting

      Predicts future by analyzing the past and present data. Type of Forecasting methods are:

      • Qualitative Methods
      • Quantitative Methods
      • Average Method
      • Naive Method
      • Drift Method

2. Unsupervised machine Learning

   Unlike supervised machine learning algorithms, examples and answers are not given in advance. In this, the algorithm machine itself has to predict based on the data. Therefore, algorithms learn from test data or real data, which have not already been labeled, categorized or classified.

   In unsupervised machine learning, it identifies similar data sets and each new piece of data gives an output based on the presence or absence of such similarities.

   1. Clustering

      Grouping of similar dataset is done based on defined criteria, by identifying pattern in each dataset.

      • Hierarchical Clustering
        1. Agglomerative clustering
        2. Divisive clustering
      • K-Means Clustering
      • DBSCAN Clustering
      

   2. Dimension reduction

      In dataset, numbers of input variables (dimensionality) are reduced. Instead of using one best algorithm for all cases, it explores range of dimensionality reduction algorithm.

3. Semi-supervised machine Learning

   Semi-supervised machine learning, this algorithm lies in between both supervised machine learning and unsupervised machine learning. Because for learning, this algorithm uses both labeled and unlabeled data. The system which uses this algorithmic method is able to improve its learning ability very easily over time.

4. Reinforcement machine learning

   Reinforcement machine learning is a learning method that interacts with the environment around it by presenting actions and detects errors as well. The speciality of this algorithm is to find and detect trial and error. This method helps the machine and S/W agents to automatically detect the activities of a particular instruction, thereby further improving the performance of the system.

Benefits of machine learning

1. Identifies trends and patterns easily.
2. No human intervention needed (automation).
3. Since machine learning has the ability to learn on its own, improves the algorithm continuously over time.
4. Handles multi-dimensional and multi-variety data.
5. Wide Applications.

Drawbacks of machine learning

1. Huge volume of datasets are required to train on.
2. Considerable amount of time is required to achieve accuracy.
3. Choosing correct algorithm to get required results.
4. If small amount of not inclusive data set is used to train algorithm then it is highly susceptible to errors.