Overview of Machine Learning: Concepts and Procedures Explained

Machine learning aims to find a function that predicts outcomes based on input data, balancing complexity to avoid underfitting and overfitting.

### Understanding Machine Learning: Concepts, Processes, and Techniques
* 🤖 Machine learning is a core technology within artificial intelligence, primarily focused on how AI **mimics human intelligence** through mathematical expressions and functions.
* 📊 The relationship between input (x) and output (y) can be expressed mathematically as **y = f(x)**, where the function f represents the **decisions AI makes based on environmental data**.
* 🔍 To find this function f, two approaches can be taken: **rule-based systems** with predefined rules, and **data-driven systems** that rely on collected data to learn patterns.
* 📈 Data is fundamental to machine learning, as it consists of **observed random samples from a larger population**, which is necessary for **estimating relationships between variables**.
* 📉 The process of learning involves **minimizing errors in predictions**, typically through training data that helps develop a model, followed by evaluation against unseen data to **ensure accuracy**.
* ⚖ Model complexity plays a crucial role; a balance must be struck between **underfitting (too simple)** and **overfitting (too complex)** to achieve **optimal performance** on both training and evaluation datasets.
* 📚 Machine learning can be categorized into three types: **supervised learning** (with known outputs), **unsupervised learning** (without specified outputs), and **reinforcement learning** (focused on strategic decision-making).
* 🔄 The machine learning process follows a structured approach, often represented by the **CRISP-DM framework**, which includes **problem definition**, **data understanding**, **preparation**, **modeling**, **evaluation**, and **deployment** in practical applications.

### Overview of Machine Learning and Its Progression
Machine learning, a core technology of artificial intelligence, relies on **mathematical concepts** to mimic human intelligence. The process involves receiving environmental data (x) and making judgments to determine actions (y), which is mathematically expressed as **y = f(x)**. Data-driven artificial intelligence is essential for finding the function f, as it gathers **observed random data** to make informed decisions, with machine learning requiring a **sample from a larger population** for effective analysis.

### The Nature of Data and Learning Objectives
The data we observe may appear fixed, but it is actually **variable and obtained by chance**. The goal of learning is to identify **fundamental properties from the entire population**, rather than just the observed data. This involves finding a function that represents the **relationship between input and output variables**, with the aim of **estimating the true relationship from a finite random sample**.

### Finding the Best Model: F Hat
To determine the best F hat, we aim to **minimize the error** between the true y and the estimated y hat. This is done using training data to find the model, but measuring the model's effectiveness is **challenging without observing the entire population**. To address this, an independent evaluation dataset is created to **assess the model's performance**, with the training data collected from the past and the evaluation data from a future point in time.

### The Importance of Model Complexity in Data Analysis
Data is gathered from the past and divided into two parts for model training, with one part treated as **future data**. Model complexity refers to the range of patterns a model can express, with **more complex models able to represent a variety of patterns**. While a complex model may perform better on training data, there is an optimal model that **balances complexity and simplicity for better overall performance**.

### The Impact of Model Complexity on Data Classification
A simple linear model struggles to classify complex data, leading to **errors on both sides of the boundary**. Increasing the model's complexity helps reduce errors for both training and evaluation data, but too much complexity can lead to **overfitting**, resulting in **poor predictions for new data**. The goal is to find a model with the **right level of complexity** to avoid **underfitting and overfitting**.

### Understanding Machine Learning: Types and Goals
Machine learning aims to find the relationship between a population's **true data** and its **outcomes**, using a training dataset for model selection. There are three main types of machine learning: **supervised learning** (with provided outcomes), **unsupervised learning** (focusing on input patterns), and **reinforcement learning** (strategically selecting data to achieve desired outcomes). Reinforcement learning differs from the other two by aiming to **derive the input that leads to the desired output**, rather than just finding the function that predicts it.

### The Procedure of Machine Learning
Machine learning can be categorized into three types, with the process starting by **defining the dependent variable y**. Data is then collected from existing databases to influence y, followed by **exploratory data analysis** to understand the data's characteristics. After preprocessing the data, it is split into **training and evaluation sets**, from which candidate models are selected and tested to ultimately **choose the best model for evaluation**.

### The CRISP-DM Process in Data Analysis and Machine Learning
Data analysis concludes with the development of an **AI service**, followed by practical application and user evaluation. The CRISP-DM framework, developed by IBM, guides the data mining process, emphasizing **business understanding** and **data collection**. The process involves **data preparation**, **exploratory analysis**, **modeling**, and **final evaluation**, with the goal of **deploying the model in the field** and **iterating on its development** if necessary.

LiveWiki

데이터 학습과 지능 1 2

데이터 학습과 지능 1 2