12 Core Concepts of Data Science

Data science is a branch of science that applies the scientific method to data to study the relationships between different characteristics and draw meaningful conclusions from these relationships. Data is therefore the key element of data science.

Here are the 12 concepts of data science:

  1. Dataset
  2. Data Wrangling
  3. Data Visualization
  4. Outliers
  5. Data Imputation
  6. Data scaling
  7. Principal Component Analysis (PCA)
  8. Linear Discriminant Analysis (LDA)
  9. Data Partitioning
  10. Supervised Learning
  11. Unsupervised Learning
  12. Reinforcement Learning


A dataset is a specific instance of data that is used at a specific point in time for analysis or to build models. A data record is available in different forms, for example, numeric data, categorical data, text data, image data, voice data, and video data. A dataset can be static (not changing) or dynamic.

Data Wrangling

Data management is the process of transforming data from its raw form into an ordered form ready for analysis. Data management is an important step in data pre-processing and includes various processes such as data import, data structuring, data cleaning.

Data visualization

Data visualization is known as the one of the most important branches of data science. It is one of the most important tools for analyzing and studying the relationships between different variables. Data visualizations (eg scatter plots, line plots, bar graphs, histograms, QQ plots, smooth densities, box plots, pair plots, heat maps, etc.) can be used for analysis.

Data visualization is also used in machine learning for data pre-processing and analysis, feature selection, model creation, model testing, and model evaluation.


An outlier is a data point that is completely different from the rest of the data set. Outliers are usually just bad data, for example, due to a faulty sensor; contaminated experiences; or human error in recording data. Sometimes outliers can indicate something real, such as system malfunction.

Outliers are very customary and are contemplating in large datasets. A common way to identify outliers in a dataset is to use a box plot.

Data Imputation

 Most records contain missing values. The easiest way to deal with missing data is to simply throw away the data point. However, deleting samples or removing entire resource columns is simply not possible because we can lose so much valuable data.

One of the most common interpolation techniques is average imputation, in which we simply replace the missing value with the average value of the entire resource column. Other options for imputing missing values ​​are median or more frequent (mode), with the latter replacing the missing values ​​with the most frequent values.

Regardless of the imputation method used in your model, remember that imputation is only an approximation and therefore can lead to an error in the final model. If the data provided has already been pre-processed, you need to know how the missing values ​​were taken into account.

Data scaling

Scaling your resources will help improve the quality and predictability of your model. For example, suppose you want to build a model to predict the credit quality of a target variable based on predictive variables such as income and creditworthiness.

As credit scores range from 0 to 850, while annual income can range from $25,000 to $500,000 without scaling its characteristics, the model is skewed in favor of the income characteristic. This means that the weighting factor associated with the income parameter is very small, which results in the predictive model that predicts credit quality based solely on the income parameter.

Principal Component Analysis (PCA)

Large datasets with hundreds or thousands of resources often result in redundancy, especially when resources are correlated with each other. Sometimes, priming a model on a big dataset with several features can lead the way to over-fitting (the model captures real and random effects).

Principal Component Analysis (PCA) is a statistical method of resource extraction. PCA is used for large, correlated data. A PCA transformation achieves the following results:

  1. a) Reduce the number of features to be used in the final model by concentrating only on the components that make up most of the variance in the data set.
  2. b) Remove correlation between features.

Linear Discriminant Analysis (LDA)

LDA is linear data preprocessing transformation technique that is often used for dimensionality reduction to select relevant properties that can be used in the final machine learning algorithm.

PCR is a standalone algorithm used for resource extraction on correlated and large-scale data. The objective of LDA is to find the resource subspace that improves class detachability and reduces dimensionality. Therefore, LDA is a supervised algorithm.

Data Partitioning

In machine learning, the dataset is often divided into training and testing sets. The model is trained with the training dataset and then tested with the test dataset.

The test dataset, therefore, acts as an invisible dataset that can be used to estimate a generalization error (the error that is expected if the model is applied to an actual dataset after the model is deployed).

Supervised Learning

They are machine learning algorithms that perform training by examining the relationship between the characteristic variables and the known target variable. Supervised learning has two subcategories:

  1. a) Continuous target variables
  2. b) Discrete target variables

Unsupervised Learning

In unsupervised learning, we are dealing with untagged data or data with an unknown structure. Unsupervised learning techniques allow us to examine the structure of our data to extract meaningful information without the aid of a known outcome variable or reward function. K-means clustering is also an example of an unsupervised learning algorithm. To master your skills visit Data Science Training in Noida

Reinforcement Learning

In reinforcement learning, the objective is to develop a system (agent) that improves its performance based on interactions with the environment. Since information about the current state of the environment often includes a so-called reward signal, we can think of reinforcement learning as an area of ​​supervised learning.


Leave a Reply

Your email address will not be published. Required fields are marked *