Mean:

The average or central value

Mean Absolute Error:

As the name says it is Mean of Absolute Error. Absolute Error is the difference between forecasted value and actual value. So MAE is the mean of those values.

Example: If a weighing scale shows 10KG whereas the actual weight is 9.5, there is an Absolute Error of 0.5. Repeat the same for 10 times and take the average of the error which is MAE.

Mean Squared Error:

This is mean of average of squares of the error. This is give us an indication on how close we are to the line of best fit for the existing data.

Median:

The is the middle value

Mode:

Most frequently occurring value.

Model:

In Machine Learning, Model is a trained algorithm.

Example: Linear Regression is an algorithm which is common for predicting continuous values. But once the algorithm is trained with data it becomes a MODEL. Sales Forecasting Model is different from Housing Price forecasting model though both use Linear Regression algorithm.

Naïve Bayes classifier:

This is a classification algorithm which works on conditional probability or Bayes Theorem to classify objects. It is probability of occurrence of an event based on past experience.

Example can be marking emails as Spam or not.

Outlier:

This is something that is abnormal from the remaining data.

Example: An orange in a basket of apples is an outlier or An Apple in a basket of vegetables is an outlier.

Overfitting:

This refers to a model that has trained too well.

Example: An athlete practices 50mtrs sprint for 100 days. Chances of his success are very high only 50mtrs race but he had to participate in a 100 or 200mts race, his chances are very slim. This is because he has been trained too well only for 50mtrs.

Quartile:

Quartiles are values that divide data into 4 equal parts or quarters.

Random forest:

This is a machine learning technique that develops multiple decision trees analyzing sets of variables. This is a like a divide-and-conquer method where it randomly picks different sets of variables and samples and decision trees are applied to them. It will repeat the process and make a final prediction.

Regression:

It is a technique to understand the relationship between a dependent variable and its independent variables. This is used for forecasting problems.

Reinforcement Learning:

It is method of machine learning which learns by making mistakes. It mimics the human behaviour of learning by goal – oriented interaction with the environment.

Root Mean Squared Error (RMSE):

This is simply the square root of Mean Squared Error and is used to find out how concentrated the data is around the best fit line. This is preferred method to validate a model as it is sensitive to outliers whereas MSE is not.

Standard Deviation:

This is a value by how much the value of a group differs from mean value of the group.

Stratified sampling:

It is a probability sampling technique to avoid bias.

Example: Let us say we are to survey a class of working people in a city who are distributed as 60:40 male to female ratio. When we pick a sample of 1000 people, stratified sampling makes sure we have 600 males and 400 females in the survey group. If either of the numbers are more of less will have our survey results biased.

Supervised learning:

This is a machine learning task where the algorithm is provide with a labelled training data. This means it has the predictor variables and the target (result) variables defined.

Example: You are given a Maths problem in a classroom, once the student completes it the teacher verifies the answer. So we have the data and solution given and training is done on the data and the solution is tested against the already given answers.

Support vector machine (SVM):

SVM is a method which identifies a new sample either belong to one or other categories. It does this by assigning every observation in training set a category and builds a model on that. Example: Identifying happy and sad faces in a group of pictures.

Time series data:

As the term says we will be dealing with data that has Time in it like minutes, hours, days, weeks, months or years. Example weather forecasting is a time series data, predicting stock market for the next week is a time series data.

Unsupervised learning:

This is a type of machine learning where there is only data but no output variables.

Example: Same example as in Supervised learning but you are given only the data. Once you solve the Math problem, you do not have an output to verify against. So the algorithm is left on its own to device a solution for the problem.

Variance:

This is how far a set a numbers are spread out from the mean value.

Example: In a class of 10 people height of each student is 5, 4.8, 5.1, 5.2, 5.4, 5.6, 5, 5.3, 5.1, 4.8 respectively. The average (mean) height is 5.13 and the data is spread out. But if the height of all the 10 students is 5.2ft then the variance is 0 and the data is not spread out.

**-Hari Mindi**