Data Feature Selection and Missing Values

Replace missing values with a specific value, such as '0' for numerical columns or the most frequent value (mode) for categorical columns.

Replace missing values with the average (mean) of the existing values in the column. Suitable for numerical columns with a normal distribution.

Replace missing values with the most frequent value (mode) in the column. Useful for categorical columns.

Values that are absent from a dataset. Represented by blanks or NaN (Not a Number) in Pandas.

Data is missing randomly and there's no relationship between its absence and other observed data. Like a coin flip, any value could be missing.

The reason for missing data can be explained by other variables in the dataset. There's a pattern, but only within specific groups.

The reason behind missing values can't be explained by observed data. There's a hidden pattern that we don't know about.

Dealing with missing values is crucial in data preprocessing, as it ensures reliable and accurate results.

Missing values can bias machine learning models, leading to inaccurate predictions.

The process of transforming raw data into meaningful features that improve model performance.

Identifying and understanding which features have the most significant impact on model predictions.

The process of selecting the most important features from a dataset for building a better model. This is a subset of features that are highly impactful for model performance.

Data is represented in a grid format where each column represents a variable known as a feature. This helps to understand the structure and characteristics of the dataset.

The number of features in a dataset determines the dimensionality of the data. High-dimensional data has many features (like text or images), while low-dimensional data has fewer features (like stock market data).

The process of transforming features into a lower dimension, reducing the complexity of the data while retaining important information. This can make data more manageable and easier to analyze.

It refers to a situation where two or more features are highly correlated, making it difficult to determine their individual impact on the output. This reduces the accuracy of the model.

Feature selection is the process of choosing and retaining relevant features, while dimensionality reduction transforms the existing features into a lower dimension, changing the data structure.

Dimensionality Reduction aims to simplify the data by reducing the complexity while retaining important information. This allows for better model performance.

Feature selection helps improve the model by removing irrelevant or redundant features that can add noise and negatively impact performance.

A technique used to address class imbalance in datasets. It creates synthetic data points for the minority class by interpolating between existing points, focusing on the feature space.

A data preprocessing step that transforms features to a similar scale. It ensures features contribute equally to the model and helps prevent bias from features with varying magnitudes.

Feature scaling often enhances the performance and convergence of machine learning models. By bringing features to a similar scale, it prevents models being dominated by features with larger values.

Machine learning algorithms that rely on distance calculations (like K-Nearest Neighbors and K-Means) benefit greatly from feature scaling. This ensures that all features have equal impact in the calculations.

Feature scaling is crucial for Principal Component Analysis (PCA) because it operates by finding the directions of greatest variance. Scaling prevents features with larger magnitudes from influencing the results disproportionately.

A common practice in machine learning involves scaling features before building a model. This is typically performed after data partitioning to avoid introducing bias.

Feature scaling helps achieve better model performance, faster convergence, and reduces bias. It transforms features into a comparable scale, leading to more reliable and accurate results.

These are methods that use the relationships between features to decide which ones to include in the model. Think of it as finding the strongest connections between ingredients and the final dish.

Methods that use the performance of a machine learning model to guide the selection of the most important features. It trains the model with different feature sets, seeing which combination results in the best outcome.

A statistical method that assesses the relationship between a numerical input variable and a binary categorical output variable. It helps identify if there is a significant association between the input and the outcome.

A statistical method used to assess the relationship between a numerical input variable and a multi-class categorical output variable. It helps determine if there is a significant difference in the outcome based on different values of the input.

A statistical method that measures the association between two categorical variables. It is used to identify if there is a statistically significant relationship between the two variables.

A statistical method that measures the strength of association between two categorical variables. It ranges from 0 to 1, with higher values indicating a stronger relationship.

A technique to deal with imbalanced datasets where you remove some observations of the majority class to create a more balanced dataset.

A technique to deal with imbalanced datasets where you add more copies to the minority class to create a more balanced dataset.

A method of under-sampling where the minority class is over-represented in the training data. This can lead to more accurate results compared to random under-sampling. The chosen minority class sample may not be a true representation of the population.

A Python package used for balancing imbalanced datasets. It offers various methods like over-sampling and under-sampling.

The advantage of random under-sampling is its ability to improve run time and storage problems by reducing the training data set. However, this can come with the potential for discarding important information.

The disadvantage of random under-sampling is the risk of creating a biased sample, which could lead to inaccurate results. This happens because the chosen sample may not accurately represent the real world data.

One advantage of over-sampling is that it avoids the loss of information that can occur with under-sampling. However, this might lead to overfitting since it replicates the minority class events.