- The end goal is to build a
model/engineand train it on some input data. The more input data, the more accurate themodelshould be - The
modelthen learns pattens in the input data - The
modelcan then do predictions based on new data you give it, with a certain level of accuracy
Common workflows include
- Import the data, commonly CSV
- Clean the data
- commonly used to remove duplicates, irrelevant or incomplete data
- you dont always remove, you could just assign a default value
- text based data needs to be represented by numerical values, example 0 for girl and 1 for boy, although in 2025 you could have -1 for cat 🐈
- Split the data into training and test sets, example 70% train and 30% test
- training input data by convention is
X - testing output data by convention is
y
- training input data by convention is
- Create model by seleting and algorithm to analyze the data
- Train the model by fitting the training data, it looks for patterns in the data
- Make predictions by passing test data and examining the output
- Evaluate and improve by measuring the accuracy of the output, if its bad, and most of the time when starting out it will be, the you need to
- create a new model with a different algorithm
- or fine tune model parameters to optimize the accuracy
There are many and certain input data / problem you're trying to solve work better with different algorithms, there are always pros and cons in terms of accuracy and performance.
- Decision trees:
- Used to classify data and predict numbers using a tree-like structure. They are easy to check and understand.
- https://scikit-learn.org/stable/modules/generated/sklearn.tree.DecisionTreeClassifier.html
- Neural networks:
- It works like the human brain with many connected nodes. They help to find patterns and are used in language processing, image and speech recognition, and creating images.
- https://www.tensorflow.org/api_docs/python/tf/keras/applications/ResNet50
- Linear regression: It predicts numbers based on past data. For example, it helps estimate house prices in an area.
- Logistic regression: It predicts like "yes/no" answers and it is useful for spam detection and quality control.
- Clustering: It is used to group similar data without instructions and it helps to find patterns that humans might miss.
- Random forests: They combine multiple decision trees to improve predictions.
This list came from https://www.tutorialspoint.com/machine_learning/index.htm
Scikit-learn has a cheat sheet https://scikit-learn.org/stable/machine_learning_map.html
- label; In supervised machine learning, the "answer" or "result" portion of an example.
- feature; An input variable to a machine learning model.
- over fitting; training performance is good but validation performance is poor
- under fitting; both training performance and validation performance are poor
Google has an extensive glossary here https://developers.google.com/machine-learning/glossary