Data scientists find themselves at the unique intersection of machine learning, statistics, and data mining. The mix helps gain new knowledge from the existing data with statistical and algorithmic analysis. In data science, a variety of machine learning algorithms gets used to solve different types of problems as a single algorithm may not be the best option for all the use cases. Data scientists commonly use various types of machine learning algorithms.
1. Linear and Logistic Regression
Linear and Logistic Regression
Linear regression
Regression analysis is a process of estimating the relationship between dependent variables. Linear regression handles regression problems, whereas logistic regression handles classification problems. Linear regression is an estimation method that’s 200+ years old.
Let’s say variable y is linearly dependent on the variable x. Regression analysis is the process of estimating the constants a and b in the equation y= ax + b. These constants express the linear relationship between the variables x and y.
Linear regression quantifies the relationship between one or more predictor variable(s)and one outcome variable. Linear regression is a popular machine learning algorithm for people new to data science. It requires students to estimate properties from their training datasets.
The Swedish Auto Insurance Dataset on Kaggle is a straightforward case study applying linear regression analysis to understand the relationships in different data sets. The case study predicts the total payment for all insurance claims, given the overall number of claims.
Logistic Regression
Logistic regression refers to a statistical procedure for crafting machine learning models where the dependent variable is dichotomous: i.e., binary. The method gets leveraged to describe data and the relation existing between one dependent variable and one or more independent variables. Coursera’s application of logistic regression analysis to predict home prices based on home-level features is a good case study to understand the algorithm.
You might also like
2. Decision Trees and Random Forests
Decision Trees and Random Forests
A decision tree implies the arrangement of the data in the form of a tree structure. Data gets separated at each node on the tree structure into different branches. The data separation happens as per the values of the attributes at the nodes. However, decision trees are prone to high variance.
Check out full articles: The Guide to Decision Tree-based Algorithms in Machine Learning (Including Real Examples)
In many machine learning algorithms with examples, you’ll see high variances, making decision tree results weak to the specific training data used. One can reduce variance by building multiple models with highly correlated trees from samples of your training data.
Bagging is the process name, and it can reduce the variance in your decision trees. Random Forest is an extended form of bagging. In addition to building trees from different samples of your training data, the machine learning algorithm also inhibits the features that can be used to construct the trees. Hence each decision tree is forced to be different.
The Institute of Physics recently published an interesting study where researchers used decision trees and random forests to predict loan default risks. Their machine learning algorithms with examples can help banking authorities select suitable people from a given list of loan applicants.
The researchers used decision trees and random forests to assess the risk associated with each potential borrower (from a list of candidates). They used both the machine learning algorithms on the same dataset. The researchers concluded that the Random Forest algorithm provided more accurate results than the Decision Tree algorithm.
Learn more our project Random Forest in Action: Predicting Climate Change and Forced Displacement
3. Gradient Boosting Machines
Gradient boosting machines like XGBoost, LightGBM, and CatBoost are the go-to machine learning algorithms for training on tabular data. XGBoost is easier to work with as it’s transparent, allows the easy plotting of trees, and has no integral categorical features encoding.
Researchers at the Center for Applied Research in Remote Sensing and GIS (CARGIS) in Vietnam recently used the three robust boosting machines (XGBoost, LightGBM, and Catboost) in combination with a Convolutional Neural Network (CNN) to classify land covers.
The research proved that combining CNN-based gradient boosting machine learning algorithms with object-based image analysis can create a more accurate method for land cover analysis.
Real-World Course (Including XGBoost): Detecting Heart Disease with Ensemble Machine Learning Methods
Build your portfolio with real-world projects from Omdena
4. Convolutional Neural Networks (CNN)
CNN features the types of machine learning algorithms that categorize images into labeled classes. The different layers of the CNN extract image feature from data sets. Gradually, they learn to classify the images.
Original Network Structure from Traffic density estimation method from small satellite imagery: Towards frequent remote sensing of car traffic
CNNs in Action: Saving Lives on Roads
Omdena recently published a case study where CNNs were used to increase road safety. Researchers used pre-trained CNNs to count and classify vehicles on roads. The algorithms also analyzed traffic flow and satellite images to create safer vehicle flow suggestions. See more here.
Visualization of the vehicle count predictions in the sample area. Source: Omdena
5. Bayesian Approaches
Naive Bayes classifiers refer to a group of classification algorithms based on Bayes’ Theorem. In a Naive Bayes classification algorithm, a class gets assigned to an element of a set with the highest probability according to Bayes’ theorem.
- Suppose A and B are probabilistic events.
- Let P (A) be the probability of A being true.
- P(A|B) stands for the conditional probability of A being true in case B is true.
Then, according to Bayes’ theorem:
P (A|B) = (P (B|A) x P (A)) /P (B)
Is the machine learning algorithms list getting a bit confusing? Don’t worry. BayesiaLab has a simple real-life case study to help you understand Bayesian networks and approaches. In the case study, Bayesian networks get used as the framework.
Bayesian approaches enable researchers to develop a faster and more economical alternative to market research. You can employ the Bayesia Market Simulator to perform market share simulations on your desktop.
6. Dense Neural Networks
Neurobiology is an inspiration for Deep Neural Networks (DNNs). They’re Artificial Neural Networks (ANN) with more layers between the input/output layers. Deep refers to the more complicated functions regarding the number of layers/units in each layer. DNNs are of three categories –
- Convolutional Neural Networks (CNN)
- Recurrent Neural Networks (RNN)
- Multilayer Perceptrons (MLPs)
Multilayer Perceptron (MLP) models are the straightforward DNN, consisting of a series of wholly connected layers. In an interesting case study from Iran, researchers estimated clay volume in a local reservoir formation based on six types of well logs using MLP networks to simulate results.
7. Recurrent Neural Networks
Recurrent Neural Networks
RNNs are a class of ANN that also use sequential data feeding. They are a type of machine learning algorithm that helps address time-series problems of serial input data to provide:
- Machine Translations
- Speech Recognition
- Language Modelling
- Text Generation
- Video Tagging
- Generate Image Descriptions
Omdena’s case study on using RNNs to make predictions of Sudden Cardiac Arrests (SCAs) based on patients’ vitals and static data is beneficial.
You can learn more about “RNNs in Action: Predicting Cardiac Arrest” here.
8. Transformer Networks
Transformer networks are neural network architectures that use attention layers as their core building blocks. A relatively new machine learning algorithm, it is revolutionizing the field of Natural Language Processing. Some famous pre-trained transformer networks include –
- BERT
- GPT-2
- XLNet
- MegatronLM
- Turing-NLG
- GPT-3
Here’s a case study where GPT-3 usage helps perform custom language tasks. Surprisingly, the language transformer model required very little training data to boost the accuracy of Electronic Health Records (EHRs).
9. Generative Adversarial Networks
Generative Adversarial Networks
Generative Adversarial Networks (GANs) are neural nets consisting of two networks – the discriminator and the generator. These two contend with each other. The generator generates data sets, and the discriminator validates these data sets. AI startup Spacept recently partnered with Omdena to build deep learning Generative Adversarial Networks to identify trees.
GANs for Good: Detecting Wildfires
The model was designed to prevent forest fires. Omdena’s team applied GANs to perform data labeling and data augmentation. Ultimately, the deep U-Net model could detect trees from large datasets.
Generating Images with Just Noise using GANs to Detect Trees and Wildfires
10. Evolutionary Approaches
Last on the machine learning algorithms list is a category of evolutionary optimization algorithms known as “Evolutionary Approaches” or simply EAs. Some popular algorithms in the evolutionary computation field include:
- Genetic algorithms (GA)
- Genetic Programming (GP)
- Differential Evolution (DE)
- The Evolution Strategy (ES)
- Evolutionary Programming (EP)
Here’s a case study where EAs usage to optimize warehouse storage processes. The model developed by the University of Žilina helps optimize the workloads of warehouse workers.
Ready for Real-World AI?
Omdena runs AI Projects with organizations that want to get started with AI, solve a real-world problem, or build deployable solutions within two months.
If you want to learn more about us, you can check out all our projects and real-world case studies here.
