Statistical Analysis and Machine Learning Fundamentals

Posted on Jun 17, 2026 in Mathematics

Measures of Central Tendency and Dispersion

a) Measures of Central Tendency

These metrics identify the central point of a data distribution.

  • Mean (Average): The sum of all values divided by the count.
    • Example: A retail store counts daily customers over 5 days: [10, 15, 20, 25, 30]. The Mean is (10+15+20+25+30)/5 = 20.
  • Median (Middle Value): The middle number in a sorted list. It is highly resistant to outliers.
    • Example: If software engineer salaries are [$60k, $65k, $70k, $80k, $250k], the Median is $70k. (The Mean would be skewed to $105k by the single high earner).
  • Mode (Most Frequent): The most common value in the dataset.
    • Example: In a shoe store, the sizes sold are [7, 8, 8, 8, 9, 10]. The Mode is 8.

b) Measures of Dispersion

These metrics quantify how spread out the data points are.

  • Range: The gap between the highest and lowest values.
    • Example: If a stock’s highest price in a day is $150 and its lowest is $142, the Range is $8.
  • Variance: The average of the squared differences from the Mean. It measures the internal volatility of the data.
    • Example: In a production line measuring item weights, a low variance means weights are highly consistent; a high variance means weights fluctuate wildly.
  • Standard Deviation (SD): The square root of variance, returning the dispersion metric back to the original unit of measurement.
    • Example: If the mean class test score is 75% with an SD of 5%, most students scored between 70% and 80%.

Skewness, Kurtosis, Correlation, and Regression

a) Skewness

Skewness defines the lack of symmetry in a frequency distribution.

  • Positive Skewness (Right-Skewed): The long tail extends to the right. The mean is pulled toward the higher values.
    • Example: Wealth distribution. A small percentage of billionaires creates a long right tail, while most citizens cluster at the lower end.
  • Negative Skewness (Left-Skewed): The long tail extends to the left. The mean is pulled toward lower values.
    • Example: Age at natural death. Most individuals pass away at an older age (creating a peak on the right), while fewer die at a very young age (creating the left tail).

b) Kurtosis

Kurtosis measures the “tailedness” or the sharpness of the peak of a data distribution compared to a normal distribution.

  • Leptokurtic (High Kurtosis): The distribution has a sharp, tall peak and heavy, thick tails. This indicates a high concentration of data around the center alongside an increased likelihood of extreme outliers.
    • Example: Daily stock market returns. Prices usually fluctuate very little (sharp central peak), but market crashes cause extreme sudden drops (heavy tails).
  • Platykurtic (Low Kurtosis): The distribution features a flat, broad peak and thin tails. The data is spread out more evenly.
    • Example: The probability of rolling a fair six-sided die multiple times—the outcomes are uniformly distributed across the board.

c) Correlation and Regression

  • Correlation: Measures the strength and direction of a relationship between two variables. It goes from -1 to +1.
    • Example: The correlation between study hours and exam scores is close to +0.8 (Strong positive relationship). It does not mean study hours caused the score (other factors like sleep matter), only that they move together.
  • Regression: An equation that models the mathematical dependency of a target variable on one or more predictor variables.
    • Example: Real estate pricing. A regression equation might look like: This lets you mathematically calculate a house value based on its size.

Machine Learning in Real-Life Applications

Healthcare

  • Use Case: Medical Imaging and Early Cancer Detection. Computer vision models are trained on millions of historical X-rays and MRI scans to recognize malignant cellular patterns.
  • Benefits: Speeds up diagnostic times drastically and spots micro-anomalies that a fatigued human radiologist might overlook.
  • Challenges: Data privacy laws make it hard to source diverse medical training data. Furthermore, models struggle with a lack of interpretability (the “black box” problem)—doctors must know why an AI flagged an image before performing surgery.

Finance

  • Use Case: Credit Scoring and Algorithmic Fraud Detection. Natural Language Processing (NLP) and anomaly detection engines scan transactional data lines instantly.
  • Benefits: Minimizes banking losses by flagging compromised credit cards instantly. It also allows unbanked individuals to get loans via alternative data evaluation (e.g., utility bill payment histories).
  • Challenges: Models can inherit historical human bias, leading to unfair credit rejections for certain demographic groups. Financial models must also adapt to changing fraud tactics.

Self-Driving Cars

  • Use Case: Autonomous Navigation. Deep neural networks process real-time feeds from LiDAR, radar, and cameras to segment road boundaries, track pedestrians, and predict vehicle trajectories.
  • Benefits: Eradicates the primary cause of traffic accidents—human errors like distracted driving, intoxication, or slow reaction times.
  • Challenges: Managing edge cases (unpredictable scenarios like a pedestrian in a costume jaywalking during a heavy snowstorm). There are also complex ethical and legal liabilities when an accident occurs.

Classification in Machine Learning

Classification is a supervised learning task where the model predicts a discrete categorical class label for a given input.

Types of Classification

  1. Binary: Two options (e.g., [Pass / Fail]).
  2. Multiclass: More than two options, but the input belongs to only one class (e.g., sorting mail into [Finance], [HR], or [Marketing]).
  3. Multi-label: An input can belong to multiple classes at once (e.g., an article tagged as both [Politics] and [Economy]).

Core Classification Algorithms

  • Logistic Regression: Despite its name, it is a classification baseline. It outputs a probability value between 0 and 1 using a sigmoid function, mapping inputs to a binary target.
  • K-Nearest Neighbors (KNN): Classifies a new data point based on the majority vote of its closest neighbors using distance metrics.
  • Support Vector Machine (SVM): Finds a hyperplane decision boundary that maximizes the margin distance between classes.
  • Decision Trees / Random Forests: Uses flowchart-like split rules to isolate data. A Random Forest combines multiple individual decision trees to generate a more stable, accurate consensus vote.
  • Naive Bayes: A probabilistic classifier based on Bayes’ Theorem that assumes all input features are independent of each other.

Regression Analysis in Machine Learning

Regression is a supervised learning task where the model predicts a continuous, numerical value.

Simple & Multiple Linear Regression

Assumes a straight-line relationship between the independent input variables (X) and the dependent numerical output variable (Y).

  • Formula: Y = mx + b
  • Example: Estimating a car’s fuel efficiency (Y) based on its engine size (X_1) and vehicle weight (X_2).

Logistic Regression

Though structurally built on linear regression principles, it passes its output through the Sigmoid Function to clamp the final value between 0 and 1. It is used to calculate probabilities for categorical classification tasks.

  • Formula: P = 1 / (1 + e^-z)
  • Example: Predicting the probability (e.g., 85% chance) that a patient has a specific medical condition based on blood markers.

Polynomial Regression

Used when the relationship between the data points is non-linear. It adds higher-degree polynomial terms (X^2, X^3) to fit curved distributions.

  • Formula: Y = a + bX + cX^2
  • Example: Tracking the growth rate of an epidemic over time, which curves upward exponentially before flattening out.

Steps Involved in Preparing a Machine Learning Model

Developing a robust machine learning model follows a rigorous lifecycle:

  1. Define Goals: Establish clear business objectives and performance metrics. Decide whether the problem is a classification, regression, or clustering task, and define success (e.g., “Achieve over 95% classification accuracy”).
  2. Data Exploration (EDA): Use summary statistics and visual charts (like histograms and scatter plots) to understand the dataset’s structure, identify correlations, and check for missing values.
  3. Data Cleaning & Preprocessing: Prepare the raw data for training. This involves handling missing data via imputation, removing duplicate records, scaling numerical values, and encoding categorical variables into numeric formats.
  4. Model Validation: Split the dataset into distinct subsets (e.g., using the Holdout method or K-Fold Cross Validation). This ensures the model is trained on one portion of the data and validated on another to check how well it generalizes to unseen data.
  5. Model Optimization: Fine-tune the model to improve performance. This includes hyperparameter tuning (using techniques like Grid Search or Random Search) and applying regularization to prevent overfitting.
  6. Deployment & Monitoring: Push the finalized model into a live production environment (via APIs or cloud services) to serve real-time user requests. The model must be continuously monitored to ensure its accuracy does not degrade over time due to data drift.