Machine Learning Basic Introduction

What is Machine Learning??

Machine learning is way of “Statistical learning”. It is a study of Statistical Algorithms in programming language which can understand by Computers. That’s why we treat these as Computer Algorithms which can improve their experience day to day by use of data. It detects patterns in data & learn how the patterns are to make predictions & recommendations through the data by processing the data.??

Machine Learning provides Predictions & Prescriptions.?

Types of Analytics??

1. Descriptive: Describe what happened.?
2. Predictive: What will happen.?
3. Prescriptive: What to do to reach goals.?

Types of Machine Learning?

1. Supervised Learning?
2. Unsupervised Learning?
3. Reinforcement Learning?

Supervised Learning:?

An Algorithm which uses training data to learn the relationship of given inputs to a given output. We use supervised learning When you need an algorithm to calculate how to classify the input data & the type of behavior you want to predict.?

Unsupervised Learning:?

An Algorithm explores variables without having any Target variable. We don’t classify the data here in Unsupervised Learning. The algorithm itself find patterns & classify the data for you.?

Reinforcement Learning:?

An algorithm learns to perform a task simply by trying to maximize result it receives for its computation. We don’t have lots of training data here. The way to learn about the environment is to interact with it.??

?

Supervised Learning Again have 2 types.?

1. Regression Analysis?
2. Classification?

Regression Analysis we will use when target variable is numeric continuous. We will use Regression in finance & investment to value assets & future prediction of value of the properties or Stocks Etc. We use regression in CAPM (Capital asset pricing model). In various companies Regression is used by managers to forecast sales & employee retention or recruiting the best people.?

?

Classification we will use when the target variable is Numeric Discrete or Categorical ordinal or Nominal. Classification categorizes the object belongs to which class among various classes of the target variable. Basically, Classification is a Predictive Modeling. Mostly classification is used in medical sector & Transport Planning sector to get accuracy of the tasks they have done.??

?

Advantages of Supervised Learning:?

1. Supervised learning gives output from the previous experience.?
2. It optimizes performance criteria by using previous experience.?
3. It always helps you to solve real world computation problems.?

Disadvantages of Supervised Learning:?

1. A lot of computation time & Memory.?
2. Classifying large data is real challenge.??
3. Decision boundary is over trained if we don’t have sufficient data.?

?

Unsupervised Learning have 2 types?

1. Clustering?
2. Dimensionality Reduction?

?

Clustering is forming a group of data points having similar qualities in a large group of datasets. Simply forming a group of similar quality items. Clustering has 2 types. In hard clustering a datapoint belongs to one cluster only where in soft clustering the pattern provided is a probability likelihood of a data point which belongs to each cluster. In anomaly detection, Genetic research widely using clustering.?

?

Dimensionality Reduction means reducing the features of a dataset. When there are large number of features the model complexity increases. That’s we called as curse of dimensionality. When we are trying to get arbitrary function with certain accuracy the number of dimensions required for the estimate grows exponentially. This causes for more sparsity which means the feature having a zero value. If any data having more sparse features the time of execution & space for the data increases. In email classification widely using this technique.??

?

Advantages of Unsupervised Learning:?

1. No previous knowledge required.?
2. Human error is minimized.?
3. Easy & fast to carry out.?

?

?

?

Disadvantages of Unsupervised Learning:?

1. No need to represent features on ground.?
2. It won’t consider spatial relationship in the data.?
3. Time taking in interpret the spectral classes.?

?

Reinforcement Learning have 2 types?

1. Positive Reinforcement Learning?
2. Negative Reinforcement Learning?

When a task is getting strength & frequency increasing due to behavior of an event occurs it is called Positive Reinforcement Learning.??

When a negative condition is stopped or avoided the behavior is strengthen that is called Negative Reinforcement Learning.??

?

Advantages of Reinforcement Learning:?

1. It solves very complex problems which cannot be solved by conventional techniques.?
2. It is preferred to achieve long term goals.?
3. The learning in here very similar to human beings.?
4. The model itself correct the errors.?

Disadvantages of Reinforcement Learning:?

1. As a framework it is not accepted.??
2. Too much learning here can lead to states overload which can deviate results.?
3. Needs lots of space & time to compute.?