# Machine Learning 101 – TechFunnel

## How does Machine Learning Work?

There are basically 4 steps in developing a ML model or application. They are:

### Step 1: Choose and arrange your training data set

Teaching data is a data set representative of the information to be ingested by the machine learning application to solve the challenge is built to fixed. In certain situations, the teaching data is labeled data – designed to select classifications and features that the model will have to recognize. Other data sets are unlabeled; thus the model will have go remove those characteristics and allocate categorizations on its own.

Nonetheless, the teaching data must be adequately prepared and scanned for anomalies or falsities that could affect the training. It should be categorized into 2 subsets: the teaching subset, which will be utilised to teach the model, and the analysis subset, used to evaluate and enhance it.

### Step 2: Select an algorithm to operate on the teaching data set

The type of algorithm is determined by the type (whether labeled or unlabeled), the quantity of data in the teaching data, and the kind of problem to be fixed. Below are the common types of ML algorithms to be utilised when labeled data:

- Recession algorithms (such as linear and logistic regression, as well as a support vector machine).
- Decision trees
- Instance-based algorithms
- Unlabeled data uses the following algorithms:
- Clustering algorithms
- Association algorithms
- Neural networks

### Step 3: Teaching the algorithm to build the application

Teaching the algorithm is a crucial process, involving operating variables via the algorithm, making comparison between the output and the outcomes it should have produced, adjusting biases and weights within the algorithm which might generate a more accurate outcome, and testing the variables again till the algorithm delivers the desired outcome most of the time. The eventual trained, precise algorithm is the machine learning application.

[embedded content]

### Step 4: Utilizing and refining the application

The last step is using the application with fresh data so that it can increase effectiveness and accuracy over time. The source of the new data will be determined by the problem being solved. For instance, **machine learning applications **built to detect spam will ingest email messages, but a ML application that runs a robot vacuum cleaner will use data generated from real-world interaction with new objects or moved furniture in the room