Creating Heatmap with Python | upGrad blog

An algorithm is a set of rules or instructions that are followed by a computer programme to implement calculations or perform other problem-solving functions. As data science is all about extracting meaningful information for datasets, there is a myriad of algorithms available to solve the purpose.

Data science algorithms can help in classifying, predicting, analyzing, detecting defaults, etc. The algorithms also make up the foundation of machine learning libraries such as scikit-learn. So, it helps to have a solid understanding of what is going on under the surface. 

Machine Learning Algorithms for Data Science

Machine learning algorithms form the core of data science applications. They enable computers to learn from data and make predictions or decisions without being explicitly programmed. This section will explore various machine learning algorithms, including supervised learning algorithms like regression and classification and unsupervised learning algorithms like clustering and dimensionality reduction.

Learn data science programs from the World’s top Universities. Earn Executive PG Programs, Advanced Certificate Programs, or Masters Programs to fast-track your career.

Read: Machine Learning Algorithms for Data Science

Commonly Used Data Science Algorithms

1. Classification

It is used for discrete target variables, and the output is in the form of categories. Clustering, association, and decision tree are how the input data can be processed to predict an outcome. For example, a new patient may be labelled as “sick” or “healthy” by using a classification model. 

2. Regression

Regression is used to predict a target variable as well as to measure the relationship between target variables, which are continuous in nature. It is a straightforward method of plotting ‘the line of best fit’ on a plot of a single feature or a set of features, say x, and the target variable, y. 

Regression may be used to estimate the amount of rainfall based on the previous correlation between the different atmospheric parameters. Another example is predicting the price of a house based on features like area, locality, age, etc.

Let us now understand one of the most fundamental building blocks of data science algorithms – linear regression. 

3. Linear Regression 

The linear equation for a dataset with N features can be given as: y = b0 + b1.x1 + b2.x2 + b3.x3 + …..bn.xn, where b0 is some constant. 

For univariate data (y = b0 + b1.x), the aim is to minimize the loss or error to the smallest value possible for the returned variable. This is the primary purpose of a cost function. If you assume b0 to be zero and input different values for b1, you will find that the linear regression cost function is convex in shape. 

Mathematical tools assist in optimizing the two parameters, b0 and b1, and minimize the cost function. One of them is discussed as follows. 

4. The least squares method

In the above case, b1 is the weight of x or the slope of the line, and b0 is the intercept. Further, all the predicted values of y lie on the line. And the least squares method seeks to minimize the distance between each point, say (xi, yi), the predicted values. 

To calculate the value of b0, find out the mean of all values of xi and multiplying them by b1 . Then, subtract the product from the mean of all yi. Also, you can run a code in Python for the value of b1 . These values would be ready to be plugged into the cost function, and the return value will be minimized for losses and errors. For example, for b0= -34.671 and b1 = 9.102, the cost function would return as 21.801. 

Our learners also read: Learn Python Online for Free

5. Gradient descent 

When there are multiple features, like in the case of multiple regression, the complex computation is taken care of by methods like gradient descent. It is an iterative optimization algorithm applied for determining the local minimum of a function. The process begins by taking an initial value for b0  and b1 and continuing until the slope of the cost function is zero.

Suppose you have to go to a lake that is located at the lowest point of a mountain. If you have zero visibility and are standing at the top of the mountain, you would begin at a point where the land tends to descend. After taking the first step and following the path of descent, it is likely that you will reach the lake. 

While cost function is a tool that allows us to evaluate parameters, gradient descent algorithm can help in updating and training model parameters. Now, let’s overview some other algorithms for data science. 

6. Logistic regression 

While the predictions of linear regression are continuous values, logistic regression gives discrete or binary predictions. In other words, the results in the output belong to two classes after applying a transformation function. For instance, logistic regression can be used to predict whether a student passed or failed or whether it will rain or not. Read more about logistic regression.

7. K-means clustering

It is an iterative algorithm that assigns similar data points into clusters. To do the same, it calculates the centroids of k clusters and groups the data based on least distance from the centroid. Learn more about cluster analysis in data mining.

8. K-Nearest Neighbor (KNN)

The KNN algorithm goes through the entire data set to find the k-nearest instances when an outcome is required for a new data instance. The user specifies the value of k to be used.

upGrad’s Exclusive Data Science Webinar for you –

Watch our Webinar on How to Build Digital & Data Mindset?

9. Principal Component Analysis (PCA)

The PCA algorithm reduces the number of variables by capturing the maximum variance in the data into a new system of ‘principal components’. This makes it easy to explore and visualize the data. 

10. Decision Trees

Decision trees are intuitive algorithms that utilise a hierarchical structure of decisions and outcomes. They are often used for classification and regression tasks, enabling the understanding of complex relationships in the data.

11. Random Forest

Random Forest is an ensemble learning algorithm that combines multiple decision trees. It is known for its high accuracy and robustness, making it suitable for tasks like image classification, fraud detection, and recommendation systems.

12. Support Vector Machines (SVM)

Support Vector Machines are powerful algorithms used for classification and regression tasks. They excel in handling high-dimensional data and are widely employed in image recognition, text categorisation, and bioinformatics.

13. Gradient Boosting

Gradient Boosting is an ensemble learning technique that combines weak learners to create a strong predictive model. It is highly effective in solving complex regression and classification problems and has gained popularity in the Kaggle community.

14. Neural Networks

Neural Networks mimic the structure and function of the human brain, making them powerful algorithms for various tasks such as image recognition, natural language processing, and speech synthesis.

Apriori

Apriori is a classic algorithm in the field of data mining and association rule learning, which is widely used in data science for market basket analysis, recommender systems, and other related tasks. It is designed to discover frequent itemsets in a transactional dataset and extract meaningful associations or relationships between different items.

The Apriori algorithm takes its name from the concept of “priori knowledge,” which refers to the assumption that if an item set is frequent, then all of its subsets must also be frequent. This assumption allows the algorithm to efficiently prune the search space and reduce the computational complexity.

Here’s a step-by-step overview of the Apriori algorithm:

1. Support Calculation: The algorithm starts by scanning the transactional dataset and counting the occurrences of individual items (1-itemsets) to determine their support, which is defined as the fraction of transactions that contain a particular item. Items with support above a predefined threshold (minimum support) are considered frequent 1-itemsets. 
2. Generation of Candidate Itemsets: In this step, the algorithm generates candidate k-itemsets (where k > 1) based on the frequent (k-1)-itemsets discovered in the previous step. This is achieved by joining the frequent (k-1)-itemsets to create new candidate k-itemsets. Additionally, the algorithm performs a pruning step to eliminate candidate itemsets that contain subsets that are infrequent. 
3. Support Counting: The algorithm scans the transactional dataset again to count the occurrences of the candidate k-itemsets and determine their support. The support count is obtained by checking each transaction and identifying the presence of the candidate itemset. Once again, only the candidate itemsets with support above the minimum support threshold are considered frequent. 
4. Repeat: Steps 2 and 3 are repeated iteratively until no more frequent itemsets can be found. This means that the algorithm progressively generates larger and larger candidate itemsets until no more frequent itemsets can be discovered. 
5. Association Rule Generation: After the frequent itemsets have been identified, the Apriori algorithm can be used to generate association rules. An association rule is an implication of the form X -> Y, where X and Y are itemsets. The confidence of an association rule is calculated by dividing the support of the combined itemset (X U Y) by the support of the antecedent itemset (X). Rules with confidence above a predefined threshold (minimum confidence) are considered significant.

Advantages and Disadvantages of Apriori

The Apriori algorithm has some advantages and limitations. On the positive side, it is relatively easy to understand and implement. It also guarantees completeness, meaning that it will find all the frequent itemsets above the minimum support threshold. 

However, it can be computationally expensive, especially for large datasets, due to the potentially exponential growth of the number of candidate itemsets. Various optimization techniques, such as pruning strategies and efficient data structures, have been proposed to address this challenge.

Wrapping Up

The knowledge of the data science algorithms explained above can prove immensely useful if you are just starting out in the field. Understanding the nitty-gritty can also come in handy while performing day-to-day data science functions. 

If you are curious to learn about data science, check out IIIT-B & upGrad’s Executive PG Program in Data Science which is created for working professionals and offers 10+ case studies & projects, practical hands-on workshops, mentorship with industry experts, 1-on-1 with industry mentors, 400+ hours of learning and job assistance with top firms.