This blog post is a one stop platform for summary of different AI models that are in predominant use. The comparative assessment of these models is based on various parameters such as - Definition, Process, Main Learning Approach, Pros, Cons, and Applications. The idea is to summarize these models and make it available for a quick view. Note that the information about the model's is not limited to the contents in this post. Readers are highly encouraged to refer valid sources for additional and detailed information.
| Model | Definition | Process | Main Learning Approach | Pros | Cons | Applications |
| Linear Regression | A model that predicts a continuous output by finding the best linear relationship between input variables and target variables. | Train the model on data to find the best-fit line that minimizes the difference between the predictions and actual values. | Supervised | Simple and easy to implement, works well with linear relationships. | Cannot handle non-linear relationships, sensitive to outliers. | Sales forecasting Stock price prediction |
| Logistic Regression | A model that predicts a binary outcome by finding the best relationship between input variables and target variables using a sigmoid function. | Train the model on data to find the best-fit line that separates the positive and negative classes. | Supervised | Simple and easy to implement, works well with binary classification problems, can handle non-linear relationships using polynomial features. | Can only handle binary classification problems, sensitive to outliers. | Spam filtering Credit approval |
| Classification | A category of models that predict a class label for a given input. | Train the model on data to learn the relationship between input and output variables, then use this relationship to make categorization (binary/ multiple) of unseen data. | Supervised | Can handle multi-class problems, diverse range of algorithms to choose from. | Can overfit or underfit if not properly optimized. | Spam filtering Image classification |
| Support Vector Machines (SVMs) | A model that separates classes by finding the maximum margin hyperplane in a high-dimensional space. | Train the model on data to find the best hyperplane that separates the classes with the maximum margin. | Supervised | Can handle non-linear relationships by using kernel trick, effective in high-dimensional spaces. | Can be computationally expensive, sensitive to choice of kernel and hyperparameters. | Text classification Protein classification |
| K Means Clustering | An unsupervised learning algorithm that divides data into "K" number of clusters based on similarity. | Train the model by randomly initializing cluster centroids, then iteratively reassigning points to the nearest cluster and recalculating centroids. | Unsupervised | Easy to implement, works well with large datasets. | Can be sensitive to initial conditions and number of clusters. | Market segmentation Customer profiling |
| Association | A type of unsupervised learning that finds relationships between variables in a dataset. | Train the model by finding frequent item sets in the data, then use these to generate rules that describe the relationships between variables. | Unsupervised | Can find hidden relationships in the data, useful for market basket analysis. | Can generate too many rules or irrelevant rules. | Market basket analysis Recommender systems |
| Model-Based Reinforcement Learning | A type of reinforcement learning that models the environment and makes decisions based on these models. | Train the model by exploring the environment, updating the model based on new information, and using the model to make decisions. | Reinforcement | Can handle problems with uncertain dynamics, can plan ahead. | Can be computationally expensive, difficult to scale to complex environments. | Robotics Game AI |
| Model-Free Reinforcement Learning | A type of reinforcement learning that learns from experience without explicitly modeling the environment. | Train the model by directly mapping from states to actions, updating the policy based on rewards received. | Reinforcement | Can handle problems with unknown dynamics, can learn from experience. | Can be less sample efficient, can have stability problems. | Robotics Game AI |
| Decision Trees | A model that splits data into smaller and smaller subsets based on the values of input variables, with the goal of finding the best split that maximizes information gain. | Train the model by finding the best splits that minimize a cost function, such as entropy or Gini impurity. | Supervised | Easy to interpret, can handle non-linear relationships. | Can easily overfit, prone to high variance. | Classifying loan applicants Diagnosing diseases |
| Random Forests | An ensemble of decision trees, where each tree is trained on a random subset of the data and the final prediction is made by averaging the predictions of all trees. | Train the model by repeating the decision tree process for many trees and combining their predictions. | Supervised | Reduce overfitting, handle noisy data well, easy to interpret. | Can be slow to make predictions, can have high memory requirements. | Predicting stock prices Diagnosing diseases |
| Naive Bayes | A probabilistic model that uses Bayes' Theorem to make predictions based on prior probabilities and the relationships between input variables and the target variable. | Train the model by estimating the prior probabilities and the relationships between input variables and the target variable. | Supervised | Simple and fast to train, works well with high-dimensional data. | Assumes independence between input variables, can have limited accuracy. | Spam filtering Sentiment analysis |
| Gradient Boosting | An ensemble of decision trees, where each tree is trained to correct the mistakes of the previous tree. | Train the model by sequentially adding trees to the model, each designed to correct the mistakes of the previous tree. | Supervised | Can handle non-linear relationships, handle noisy data well. | Can be slow to train, can have high memory requirements. | Predicting stock prices Diagnosing diseases |
| Neural Networks | A model inspired by the structure and function of the human brain, consisting of layers of interconnected nodes. | Train the model by adjusting the weights of the connections between nodes to minimize a cost function. | Supervised/ Unsupervised | Can handle non-linear relationships, can learn complex representations. | Can be difficult to interpret, can be slow to train. | Image recognition Speech recognition |
| Deep Learning | A sub-field of machine learning that focuses on training deep neural networks with many hidden layers. | Train the model by training many layers of interconnected nodes to minimize a cost function. | Supervised/ Unsupervised | Can learn complex representations, can handle large amounts of data. | Can be difficult to interpret, can be computationally expensive. | Image recognition Speech recognition |
| Convolutional Neural Networks (CNNs) | A type of neural network specifically designed for image analysis, with convolutional layers that scan the image for specific features. | Train the model by training convolutional layers to identify specific features in the image and fully connected layers to make predictions based on these features. | Supervised | Effective for image analysis, can learn spatial hierarchies of features. | Can be computationally expensive, can have high memory requirements. | Image recognition Object detection |
| Recurrent Neural Networks (RNNs) | A type of neural network that processes sequential data by passing information from one step of the sequence to the next. | Train the model by unrolling the network over time, processing each step of the sequence one at a time and updating the hidden state to include information from the previous step. | Supervised/ Unsupervised | Can handle sequential data such as time-series or natural language, can handle variable-length sequences. | Can struggle with long-term dependencies, can be computationally expensive. | Speech recognition Natural language processing Time-series forecasting |