Machine Learning In Algorithmic Trading

Demystify machine learning by exploring data fitting, model families, and Python-based implementations, and learn how to apply these techniques to improve trading algorithms.

Explore how machine learning, a subset of artificial intelligence, learns from data by fitting functions to observations and letting model parameters emerge, enabling forecasts with large, high-quality datasets.

Learn about supervised learning, unsupervised learning, and reinforcement learning, with examples like regression for house prices, classification of emails, and clustering customers, plus how reinforcement learning optimizes through rewards.

Learn the building blocks of a machine learning model, from data quality and cleaning to feature selection, algorithm design, model training, evaluation, and deployment.

Explore how machine learning powers recommendation systems, autonomous vehicles, and NLP with sentiment analysis, while addressing overfitting, underfitting, data bias, and ethics in trading.

Practice supervised learning, a machine learning approach trained on labeled data where each example maps inputs to an output label, enabling price estimates for new houses and predictive modeling.

Explore key concepts in supervised learning: training data, learning algorithms that infer a function to map new examples, and the roles of regression and classification in trading.

Learn regression by predicting house prices from features such as size, bedrooms, location, and age. Build a model that learns the relationship between inputs and a continuous price.

Explore supervised learning via classification, predicting discrete categories from input features, using emails to distinguish spam from non-spam based on subject, links, body text, and sender reputation.

Learn how supervised learning trains models through data splitting, feature selection, and training, then evaluates performance with test data and metrics like accuracy and mean squared error for trading.

Explore unsupervised learning, a branch of machine learning that analyzes unlabeled data to identify underlying structures or patterns such as similarities and clustering, without labeled outputs.

Explore unsupervised learning, where models find hidden patterns in unlabeled data using clustering methods like k-means and DBSCAN to segment customers and uncover purchasing behavior.

Pre-process and clean input features for unsupervised learning, then choose the right algorithm—K-means for clustering or DBSCAN for clusters with noise—train the model to form groups, and interpret the results.

Identify the challenges and best practices of unsupervised learning, including choosing the number of clusters with elbow method, scaling, hyperparameters, and handling noise and outliers.

Explore the importance of unsupervised learning for understanding unlabeled data, including clustering, anomaly detection, and exploratory data analysis. Preview upcoming Python-based data preparation and training pipeline lessons.

Split data into training, validation, and testing sets to train models and assess accuracy on unseen data. Use the validation set to select the best model and tune parameters.

Explore three data-splitting techniques—random, stratified, and time-based splits—in the context of trading, with best practices to prevent leakage, ensure representativeness, and evaluate model performance.

Balance model complexity to optimize generalization and avoid overfitting or underfitting in trading models. Compare training and test accuracy to find the sweet spot for robust predictions on unseen data.

Learn to split data in python with random, stratified, and time-based approaches using train_test_split and time slicing on diabetes and wine data.

Explore practical machine learning classifiers—decision trees, k-nearest neighbors, random forests, logistic regression, SVM, naive Bayes—and their trading applications, with simple Python examples and use-case guidance.

The k nearest neighbors algorithm classifies a new point by majority vote among its k closest points using Euclidean distance, with k chosen by the user, illustrating instance-based, non-parametric learning.

Learn to implement a k-nearest neighbors classifier in python using sklearn, fit on two-feature data, and predict new points, including the iris data set with standard scalar and train/test split.

Explore k means, an unsupervised clustering algorithm that partitions data into k clusters by assigning points to the nearest centroid using Euclidean distance, enabling market segmentation and clustering by similarity.

Master k-means clustering in Python using scikit-learn on the iris dataset, visualize results with matplotlib, and interpret centers and labels from an unsupervised 3-cluster model.

Explore decision trees and their derivatives, like random forests and gradient boosting, for easy interpretation, robust handling of numerical and categorical data, and strong performance in noisy trading data.

Train a simple decision tree classifier with scikit-learn on a mock dataset and iris data, then predict on new data, evaluate with a 70/30 split, and visualize Gini impurity.

Apply random forests to classify or regress by combining multiple decision trees through bagging and feature subsampling, improving generalization, reducing overfitting, and revealing feature importance.

Demonstrates how to fit and predict with a random forest classifier in Python using scikit-learn, including iris data set, train-test split, and accuracy evaluation.

Logistic regression is a simple statistical model that estimates the probability of a binary outcome using a weighted sum of features and a sigmoid function, enabling threshold-based classification.

Apply logistic regression to predict the probability of passing from hours studied using a sigmoid curve, and classify by a 0.5 threshold.

Load iris data from scikit-learn, split 80/20, fit a logistic regression model, and evaluate with a classification report and confusion matrix; four features yield 100% precision, two features about 90%.

Explore how the Naive Bayes classifier uses feature independence and the product of probabilities for classification, with applications in spam filtering and text analysis.

Explore gaussian naive bayes for continuous features by modeling each class with a gaussian distribution using feature means and variances, and computing likelihoods for classification.

demonstrates a Python example with categorical and Gaussian Naive Bayes classifiers, using train-test split and a classification report to compare precision, recall, and F1 for trading data.

Explore how support vector machines use a hyperplane to separate classes, maximize margin, and extend to higher dimensions for classification, regression, and outlier detection with non-probabilistic binary decisions.

Explore how support vector machines maximize margins with soft margins, use kernels to lift data to higher dimensions, and optimize classification via cross validation.

illustrates a python SVM workflow on the iris data set with scikit-learn, using standard scaler, and comparing linear, polynomial, and RBF kernels to visualize decision boundaries in two features.

Learn to evaluate classification models in algorithmic trading using accuracy, confusion matrix, and classification report. Understand precision, recall, F1, and macro, micro, and weighted averages across binary and multiclass classifications.

Explore how the ROC-AUC and PR-AUC evaluate binary classifiers across thresholds, using true positive rate and false positive rate, and compare precision-recall with ROC in trading contexts.

Learn how to build a multiclass iris classifier with Gaussian Naive Bayes, perform a 70/30 train-test split, evaluate with accuracy, confusion matrix, and classification report, and visualize results with seaborn.

Build and evaluate a logistic regression model on synthetic binary data, compute the ROC-AUC score, and plot the ROC curve with threshold analysis and the no-skill baseline.

Explore PR-AUC in Python using a logistic regression classifier on synthetic binary data; compute precision-recall curves with predict_proba, plot with matplotlib, and achieve an area under the curve of 0.94.

Explore diverse data sources for algorithmic trading, from real-time stock data and financial statements to APIs, web-scraped news, and economic indicators, and learn labeling, data quality, and visualization.

Download historical data for trading and backtesting using two free methods: Python with yfinance and the Dukascopy platform, covering forex, majors, minors, metals, stocks, and futures.

Load daily bitcoin price data with pandas, filter out zero-volume days, plot candlesticks with Plotly, and compute EMA and RSI indicators using pandas_ta for visualization and inspection.

Explore how to visualize financial data for algorithmic trading using Python tools like yfinance, pandas, matplotlib, and Plotly to plot closing prices, candlesticks, moving averages, and indicators.

Learn how data labeling marks buy and sell moments in trading using patterns and indicators. The lesson covers labeling approaches, accuracy, biases, and cross-checking to prepare useful input data.

the fixed time horizon method labels an observation by future return over h bars against a threshold, yielding plus one, zero, or minus one; it ignores volatility and intrahorizon paths.

Demonstrate a fixed time horizon labeling method in python using label_data, which adds -1, 0, or 1 labels from future closing price changes over h bars and a threshold.

Label future price moves within the h bar horizon by comparing returns to a threshold, assigning plus one, minus one, or zero as thresholds are crossed, noting limits.

learns a python example for improved time horizon labeling in algorithmic trading by computing future threshold-based labels on euro/usd five-minute candles and visualizing results with plotly and matplotlib.

Label data with the triple barrier method using volatility-based thresholds from intraday EWMA volatility, referencing the current candle's closing price, and assign +1, -1, or 0 based on barrier outcomes.

The lecture demonstrates loading euro-usd daily data from a csv, computing daily volatility with an exponential moving standard deviation, and applying fixed horizon and volatility-based labeling, including atr labeling.

Explore dynamic, strategy-specific labeling by adjusting thresholds to match take profit and stop loss distances, applying long and short labeling with a triple bars approach and Python implementation.

Use ATR-based volatility to label long and short trading signals in a Python example. Split labels into long and short columns for clean machine learning training in algorithmic trading.

Explore optimizing strategy thresholds for long and short positions using backtesting, with a simple moving average and Bollinger bands to trigger entries, and extract best take-profit and stop-loss values.

Explore strategy optimization in Python by tuning trend signals from exponential moving averages and Bollinger band edges, with backtesting and parameter optimization for stop loss and take profit.

Optimize thresholds in a Python backtest using moving averages, RSI, Bollinger bands, and ATR; demonstrate EMR signal, grid search, and stability-focused heatmap evaluation.

Prioritize data quality and the garbage in, garbage out rule. Use indicators such as moving averages, their slope, and three moving averages alignment to reveal trend and momentum.

Derive moving averages, slopes, and distance to the EMA to label signals and prepare ML-ready data for algorithmic trading with Python.

Explore feature enhancement and dimensionality reduction techniques for algorithmic trading, including PCA, polynomial features in scikit-learn, and strategies to manage correlated features and overfitting.

Understand feature enhancement and dimensionality reduction for trading data using Python, including correlation analysis, polynomial features, and PCA with two components.

Standardize features to zero mean and unit variance, normalize to a 0-1 range when needed, and apply one hot encoding to convert categorical data for models in algorithmic trading.

Learn to standardize features with StandardScaler, apply PCA to reduce dimensions, and normalize using MinMaxScaler, then perform one-hot encoding with pandas get_dummies for categorical features in a trading dataset.

Label data, pre-process with feature reduction, train a trading classifier on historical data; apply PCA with polynomial features, compare logistic regression, SVC, and random forest using ROC and precision-recall.

Avoid data leakage by splitting time series data before pre-processing and applying transforms only on training data; compare logistic regression and boosting with threshold histograms and roc/pr metrics.

Leverage ensemble learning with xgboost, a gradient boosting framework that sequentially corrects errors from weak trees. It handles missing values with regularization and supports regression, classification, ranking, and more.

Introduce neural networks for classifiers, covering input, hidden, and output layers with weights and biases. Apply activation functions like sigmoid and use RSI and moving average slope to predict uptrends.

Explore a Python XGBoost and neural networks example for trading signals, using EMA and Bollinger Bands indicators, RSI, and a 90/10 train-test split on 10,000 rows.