Machine Learning for Business Applications Lecture Notes PDF

Machine Learning for Business Applications Introduction to ML for BA – Lecture A.0 Prof. Dr. Maximilian Schiffer Professorship of Business Analytics & Intelligent Systems TUM School of Management Munich Data Science Institute Winter Semester 2024/25 Professorship of Business Analytics & Intellige...

Machine Learning for Business Applications Introduction to ML for BA – Lecture A.0 Prof. Dr. Maximilian Schiffer Professorship of Business Analytics & Intelligent Systems TUM School of Management Munich Data Science Institute Winter Semester 2024/25 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Agenda Motivation Basics of Machine Learning Essentials & Training Strategies Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Agenda Motivation Basics of Machine Learning Essentials & Training Strategies Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München What is Machine Learning? Introduction ML for BA – Motivation Machine Learning = algorithmic decisions or predictions that are based on data Training based on historic data Application based on new data (Training phase) (Inference phase) Training Samples New Data Learning Model Model Algorithm Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 4 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Artificial Intelligence & Machine Learning Introduction ML for BA – Motivation 1950 1980 2010 Artificial Intelligence (AI) is an umbrella term for computer software that mimics human cognition in order to perform complex tasks and learn from them. Artificial Intelligence Machine Learning Deep Learning Machine learning (ML) is a subfield Engineering of Perform tasks without Machine learning of AI that uses algorithms trained on machines that mimic explicit instructions based on artificial data to produce adaptable models cognitive functions neural networks that can perform specific tasks. Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 5 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Introduction to ML - History Introduction ML for BA – Motivation 1940 - 1950 Early Days 1943: McCulloch & Pitts: Boolean circuit model of brain 1950: Turing’s “Computing Machinery and Intelligence” 1950 - 1970: Excitement 1950s: Early AI programs (Samuel’s checkers program) 1956: Dartmouth meeting: “Artificial Intelligence” adopted 1965: Algorithms for logical reasoning Using logical rules to answer questions about a specific domain of knowledge Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 6 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Introduction to ML - History Introduction ML for BA – Motivation 1940 - 1950 Early Days 1943: McCulloch & Pitts: Boolean circuit model of brain 1950: Turing’s “Computing Machinery and Intelligence” 1950 - 1970: Excitement (Samuel’s checkers program) 1950s: Early AI programs (Samuel's 1956: Dartmouth meeting: “Artificial Intelligence” adopted 1965: Algorithms for logical reasoning 1970 - 1990: Knowledge-based approaches 1974 - 79: AI Winter A system learns from an expert and 1988 - 93: Expert systems can give advices to non-experts Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 7 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Introduction to ML - History Introduction ML for BA – Motivation 1940 - 1950 Early Days 1943: McCulloch & Pitts: Boolean circuit model of brain 1950: Turing’s “Computing Machinery and Intelligence” 1950 - 1970: Excitement (Samuel’s checkers program) 1950s: Early AI programs (Samuel's 1956: Dartmouth meeting: “Artificial Intelligence” adopted 1965: Algorithms for logical reasoning 1970 - 1990: Knowledge-based approaches 1974 - 79: AI Winter 1988 - 93: Expert systems 2000 - 2020: High Performance Computing Big Data & Deep Learning Learning of intellectual tasks, e.g., large language models, image recognition… Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 8 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Introduction to ML – Overview Machine Learning Supervised Unsupervised Reinforcement Learning Learning Learning Classification Regression Clustering We teach Reinforcement e.g., can we predict who has e.g., how happy is a person e.g., which people have Learning in our Master courses a university diploma? in their current job? similar taste for movies? “Introduction to Deep Reinforcement Learning” Yes 100 15 24 and Yes Yes 67 45 “Coding Lab: Deep No 58 Reinforcement Learning” Yes No 22 29 79 32 No 0 17 … Lecture 2 & 3 Lecture 5 Lecture 4 Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 9 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Machine Learning in the Business Context Introduction ML for BA – Motivation Potential applications of ML in the business context: Fraud Detection (e.g., Visa, PayPal) Recommendations (e.g., Netflix, Youtube) Chatbots (e.g., The North Face, Decathlon) Image Generation (e.g., Netflix) Customer Segmentation (e.g., Southwest Airlines, Skyscanner) Image Recognition (e.g., BMW) Demand / Load Prediction (e.g., Walmart, Netflix, Cello) Predictive Maintenance & Predictive Supply Chain Management (e.g., Pepsi, Colgate) Personalized Marketing Campaigns with Generative AI (e.g., Persado, GetResponse) Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 10 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Machine Learning in the Business Context Introduction ML for BA – Motivation Example: Fraud Detection https://www.youtube.com/watch?v=96k0sncyoXA Classification Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 11 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Machine Learning in the Business Context Introduction ML for BA – Motivation Example: ML-based clustering of flight origins & destinations https://youtu.be/D8NlYPtPgwA?t=107s Clustering Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 12 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Machine Learning in the Business Context Introduction ML for BA – Motivation Example: Demand Prediction https://www.youtube.com/watch?v=HQ7CI7JlXJQ Regression Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 13 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Scope of the Course Introduction ML for BA – Motivation 1. Introduction to Machine Learning for Business Applications 2. Naive Bayes & Bayesian Networks Classification 3. Decision Trees 4. Clustering Clustering 5. Regression Regression 6. Neural Networks 7. Data Preparation, Generalization & Evaluation 8. Recap & Exam Preparation Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 14 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Agenda Motivation Basics of Machine Learning Essentials & Training Strategies Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München From Data to Information Introduction ML for BA – Basics of ML Interpretation & Evaluation Prediction Knowledge Selection and Preprocessing Data Consolidation Warehouse Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 16 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München From Data to Information Introduction ML for BA – Basics of ML Focus of this Prescriptive course Added Value Analytics Predictive Analytics Descriptive Analytics Complexity Descriptive Analytics Predictive Analytics Prescriptive Analytics Analysis of historical data to Use statistical models to Recommend actions to uncover trends and patterns forecast future outcomes based optimize decision-making to on past data achieve desired outcomes Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 17 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Datasets: Features and Target Variables Introduction ML for BA – Basics of ML Dataset Example: Each data points has an unordered collection of features  if the structure is essential, it needs to be embedded e.g., Feature Realization of the feature “height” Target variable Different feature types: categorical, ordinal, and numerical features Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 18 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Feature Types I – Categorical Features Introduction ML for BA – Basics of ML Categorical Features Example Each instance falls into one category of a set of categories How are categories defined? 1. Categories are mutually exclusive 2. No natural ordering of the categories 3. Thus, the only meaningful operation is equality testing ( ) What are challenges? In real-world datasets: synonyms It is crucial to analyze whether categories can be subsumed under an umbrella term Further examples of categorical data? marital status, brand of product, … Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 19 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Feature Types II – Ordinal Features Introduction ML for BA – Basics of ML Ordinal Features Example Instance falls into a set of categories with a natural ordering How are the categories treated? 1. Categories are encoded as numbers to preserve their ordering 2. It is meaningful to compare values ( ) 3. However, the “difference” between these encoded numbers should not be measured, added, or multiplied How do you differentiate ordinal from categorical data? Can be challenging sometimes. Does have a natural ordering? What are further examples of ordinal data? Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 20 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Feature Types III – Numerical Features Introduction ML for BA – Basics of ML Numerical Features Example Instances are integers or real numbers (difference relevant) How are numerical features treated? 1. It is meaningful to add, multiply, compute mean or variance (etc.) of target variables 2. Usually, target variables are normalized in order to reach Unit variance: standard deviation zero mean and unit variance: and variance of a sample will tend towards 1 as sample size tends alternatively: Normalization on [0,1]: towards infinity. 3. Before normalization, handle extreme values (outliers) e.g.: What are further examples of numerical features? Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 21 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Excursion: Normalization Introduction ML for BA – Basics of ML Normalization: rescaling numerical data to ensure that all features have similar scales and prevent one feature from dominating during training. In the following, we focus on the min-max normalization on [0,1] based on Raw Features of Houses in Neighborhood 100 # Rooms 50 0 0 20 40 60 80 100 Age Normalized on [0,1] - Features of Houses 1 # Rooms 0,5 0 0 0,2 0,4 0,6 0,8 1 Age Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 22 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Credit Scoring - Features and Target Variables Introduction ML for BA – Basics of ML Loan amount Numerical Features Instalment / disposable income Savings Ordinal Features Employment Categorical Purpose of loan Features Housing Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 23 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Three basic ML Problems Introduction ML for BA – Basics of ML Classification Regression Clustering Based on existing data, predict Prediction of continuous-valued Find a pattern in a dataset where new data belongs to output based on historic data without associated labels + + + ++ + + ++ + + + ++ ++ + + + + + ++ ++ ++ + + + + + + + + + + ++ Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 24 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Classification – Example Introduction ML for BA – Basics of ML From data with known and categorical labels, create a statistical model that determines which label to apply to a new observation Loan $? Example: Loan → yes, no Yes Yes Yes f() No decision boundary Yes Yes No Yes Yes No No No Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 25 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Classification – Notation Introduction ML for BA – Basics of ML cf. Note: We can determine the accuracy of the mapping via its error rate, i.e., the proportions of mistakes that the mapping yields Indicator function equals 1 if , 0 else Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 26 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Classification – Notation continued Introduction ML for BA – Basics of ML Data (w/o labels) Categorical Mapping Target Variable … Yes, Yes, …, No Yes No … Yes No Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 27 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Classification – Notation continued Introduction ML for BA – Basics of ML Additional differentiation: Multi-Class vs Binary Classification Multi-Class Classification - Binary Classification Classes are mutually exclusive even if the data Distinctions are one-vs-rest, i.e., let’s test this point is an outlier Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 28 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Classification – Examples and Algorithms Introduction ML for BA – Basics of ML Typical Algorithms & Example Applications Logistic Regression: Medical Diagnosis: Predicting whether a patient has a particular disease (e.g., diabetes) based on certain characteristics such as age, blood test results,... K-Nearest Neighbors (KNN): Recommendation Systems: If a user likes a particular item, then the system recommends items that similar users have liked. Decision Trees: E-commerce: Predicting whether a user will buy a product or not based on features like age, browsing history, and items in cart. Neural Networks: Image Recognition: For example, CNNs are used to automatically tag and categorize images in photo storage platforms. Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 29 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Three basic ML Problems Introduction ML for BA – Basics of ML Classification Regression Clustering Based on existing data, predict Prediction of continuous-valued Find a pattern in a dataset where new data belongs to output based on historic data without associated labels + + + ++ + + ++ + + + ++ ++ + + + + + ++ ++ ++ + + + + + + + + + + ++ Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 30 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Regression – Example Introduction ML for BA – Basics of ML From data with known and numerical labels, create a statistical model that determines which label to assign to a new observation 15 24 67 45 58 22 29 79 32 17 Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 31 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Regression – Notation Introduction ML for BA – Basics of ML cf. A specific implementation of the regression model is the (first order) linear model: Target Feature y-axis intercept residual error slope Where and are unknown and estimated from the data Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 32 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Regression – Notation continued Introduction ML for BA – Basics of ML Data Continuous Mapping Target Variable … 79, 13, …, 23 79 15 … 29 17 Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 33 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Regression – Examples and Algorithms Introduction ML for BA – Basics of ML Typical Algorithms & Example Applications Polynomial Regression: Finance: Predicting stock prices over time by modeling the relationship between historical price data and relevant factors. Random Forest Regression: Environmental Science: Estimating pollutant levels based on meteorological data, traffic patterns, and other environmental factors. Support Vector Regression (SVR): Economics: Predicting future GDP growth rates using historical economic indicators and financial data. Ridge Regression: Marketing: Modeling the relationship between advertising spending and product sales to optimize marketing budgets. Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 34 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Regression vs. Classification Introduction ML for BA – Basics of ML Non-formal distinction: Continuous target variable Regression Coarse Discrete target variable Classification The target variable being continuous or discrete steers the selection of a suitable statistical learning method for a given task. Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 35 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Three basic ML Problems Introduction ML for BA – Basics of ML Classification Regression Clustering Based on existing data, predict Prediction of continuous-valued Find a pattern in a dataset where new data belongs to output based on historic data without associated labels + + + ++ + + ++ + + + ++ ++ + + + + + ++ ++ ++ + + + + + + + + + + ++ Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 36 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Clustering – Example Introduction ML for BA – Basics of ML Identification of groupings in data, no predefined groups a priori Example clustering: which people have similar tastes in movies? Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 37 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Clustering – Notation Introduction ML for BA – Basics of ML cf. Two objectives when discovering clusters: 1) Data Similarity: Data points within the same cluster are as similar as possible 2) Cluster Separation: Data points in different clusters are as different as possible Note: 1) Unlike in classification problems, the clusters are not known a priori 2) Even to determine a sensible number of clusters is not always straight forward and may require the approximation of in a first step Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 38 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Clustering – Notation continued Introduction ML for BA – Basics of ML Dataset Clusters Mapping as Target Variables 1 1 K 2 K Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 39 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Clustering – Examples and Algorithms Introduction ML for BA – Basics of ML Typical Algorithms & Example Applications K-Means Clustering: Customer Segmentation: Grouping customers into segments based on their purchase history and behavior. Agglomerative Clustering: Social Network Analysis: Identifying communities or groups of users with similar interests or connections. Gaussian Mixture Models (GMM): Anomaly Detection: Identifying unusual patterns or outliers in financial transactions for fraud detection. Self-Organizing Maps (SOM): Data Visualization: Reducing the dimensionality of high- Annual Income [k$] dimensional data for pattern recognition. Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 40 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Agenda Motivation Basics of Machine Learning Essentials & Training Strategies Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Feature Engineering - Example Introduction ML for BA – Essentials We want to describe the state using a vector of features (properties) Features are abstract representations of states, and map states to real numbers (often 0/1) that capture important properties of the state Example features: − Distance to closest ghost − Distance to closest dot − Number of ghosts − 1 / (distance to dot)2 − Is Pacman in a tunnel? (0/1) Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 42 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Feature Engineering Introduction ML for BA – Essentials Transforming raw data into a set of meaningful and informative features that serve as input for ML algorithms. It involves selecting, creating, and transforming features to improve the predictive performance of models and enable them to capture relevant patterns and relationships in the data. ▪ Attributes are not always obvious, e.g., in the case of textual information or images. In these cases, how do we pick a representation? ▪ Feature engineering decisions depend on the problem at hand: should encode information relevant to Intuition: If and are in the same class they should have “similar” representations Similar attribute values if they are in the same class, dissimilar values if not Remaining issue: How do we measure similarity? Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 43 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Overfitting Introduction ML for BA – Essentials If we consider every anomaly when fitting the model, it overfits Results of overfitting: Model gets too large and complex Overfitting prediction error Model very targeted towards performing well on training set accuracy, but not on test sets Very hard to generalize to new data test data Potential solutions to overfitting: Model simplification Underfitting training data Early stopping More on these later in the Regularization course model complexity Pruning Ensemble methods Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 44 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Training, Validation, and Testing Introduction ML for BA – Essentials Training Set: The data used to train the model, allowing it to learn patterns and relationships. Validation Set: A separate subset of data used during training to tune model parameters and prevent overfitting. It helps improve the model by evaluating performance on unseen data Test Set: A separate subset used to assess the model’s performance after training is complete. It provides an unbiased evaluation of the model’s ability to generalize to new data. Available Data Holdout data Training Set Validation Set Test Set Often, we see the split of 60% training, and 20% each of validation and testing data – these can vary depending on the specific problem at hand Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 45 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Excursion: Cross Validation Introduction ML for BA – Essentials Cross-validation is a technique to evaluate a Example: 10-fold Cross-Validation model’s performance by splitting the dataset into multiple training and test sets. Training Set The most common form is k-fold cross- validation, where the data is divided into k Training folds Test fold subsets, and the model is trained and tested k times, each time with a different subset used as 1st iteration R1 test set. 2nd iteration R2 Process: 1. Divide data into k equal subsets. … 2. Train the model k times, each time using a different subset as the test set and the 10th iteration R10 remaining k - 1 subsets as the training set. 3. Average the results (R1 to Rk) to get a robust estimate of model performance. Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 46 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Accuracy and Un-balanced Classes Introduction ML for BA – Essentials Check: are there classes that dominate others? Example # of research affiliations ▪ You are predicting Nobel prize winners ( ) vs. not ( ) ▪ Would you prefer classification model A or B? ▪ Is the accuracy (% correct) higher for A or B? → Accuracy / error rate is a poor metric here # of academic publications Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 47 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Recap: False Positive and False Negative Introduction ML for BA – Essentials False positives are incorrect positive predictions or identifications made when the actual condition or class is negative. While false negatives are incorrect negative predictions or identifications made when the actual condition or class is positive. Prediction: Example: Actual Is it a hotdog? Condition / Class Prediction / Identification It is a hotdog True False True Positive False Positive True (correct prediction) (type I error) False Negative True Negative False (type II error) (correct prediction) Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 48 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Accuracy and Un-balanced Classes Introduction ML for BA – Essentials Check: are there classes that dominate others? Example # of research affiliations ▪ Better metrics based on true positives (TP), false negatives (FN) and false positives (FP) ▪ Precision: TP / (TP + FP) “Of all the instances the model predicted as positive, how many were actually positive?” ▪ Recall: TP / (TP + FN) “Of all the actual positive instances, how many did the model correctly predict as positive?” ▪ F1-Score: 2 * (Precision * Recall) / (Precision + Recall) harmonic mean of precision and recall # of academic publications Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 49 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Outliers in the Data Introduction ML for BA – Essentials Outliers: Isolated instances of a class that are unlike any other instance of that class. Numerical ▪ All learning methods are affected by outliers to various degrees ▪ How can you deal with outliers? ▪ Remove ▪ Identify threshold (i.e., based on confidence Categorical interval) and correct accordingly ▪ Fix (if mislabeled) Blue Red Green Yellow ▪ Visualization helps detect outliers for low dimensional data Ordinal Neutral unsatisfied Very satisfied Unsatisfied Satisfied Handling Outliers will be discussed in more detail Very later in the course Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 50 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Survivor Bias Introduction ML for BA – Essentials Definition: cognitive bias that occurs when we focus on observations that survive some process while overlooking those that did not survive, as they are no longer visible → Survivor Bias may result in wrong conclusions, e.g.: „Most castles were made out of stone.“ vs. „Most castles were built of wood, but destroyed over time“ Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 51 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München Recap Introduction Key takeaways We discussed Machine Learning (ML) and its historical developments We defined features, target variables, and different feature types We formally defined the three basic ML problems: classification, clustering, and regression with the additional differentiation between supervised and unsupervised learning In the next lecture, we will recap basic Finally, we discussed ML implementation essentials and training strategies probabilities, learn about conditional probabilities and use these concepts to classify new observations Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 52 Professorship of Business Analytics & Intelligent Systems TUM School of Management Technische Universität München References : Murphy, K.P., Machine Learning: A Probabilistic Perspective, MIT Press, 2012. : https://www.scienceabc.com/innovation/what-is-artificial-intelligence.html : https://de.m.wikipedia.org/wiki/Datei:Symbolics3640_Modified.JPG : https://www.extremetech.com/extreme/210872-extremetech-explains-what-is-moores-law : https://www.lfb.rwth-aachen.de/de/institute/facilities/it-infrastructure/ : https://medium.com/analytics-vidhya/introduction-to-convolutional-neural-networks-c50f41e3bc66 : Think like a Data Scientist – Brian Godsey : https://youtu.be/gOKjuXywNFw : https://youtu.be/D8NlYPtPgwA?t=78 : https://www.investopedia.com/terms/m/marketsegmentation.asp : https://www.youtube.com/watch?v=HQ7CI7JlXJQ : https://developers.google.com/static/machine-learning/clustering/images/NormalizeData.png : https://cyberhoot.com/cybrary/data-normalization/ : https://www.nvidia.com/en-us/glossary/data-science/recommendation-system/ : https://mikescogs20.medium.com/netflix-recommendation-system-inside-the-algorithm-55edc1712748 : https://towardsdatascience.com/customer-segmentation-using-k-means-clustering-d33964f238c3 : https://fsc.stevens.edu/dynamics-of-stock-price-reaction-to-shocks/ : https://blog.nillsf.com/index.php/2020/05/23/confusion-matrix-accuracy-recall-precision-false-positive-rate-and-f-scores-explained/ : https://ikeamuseum.com/de/erkunden/die-geschichte-von-ikea/sprichst-du-hotdogisch/ : https://www.alnatura.de/de-de/rezepte/suche/hotdogs-mit-gepickelten-zwiebeln-106031/ : https://i.pinimg.com/736x/82/56/31/825631bef898a55194d3d65ef1290a30.jpg : https://de.wikipedia.org/wiki/Pizza : https://dataschool.com/misrepresenting-data/survivorship-bias/ Winter Semester 2024/25 | Machine Learning for Business Applications | Prof. Dr. Maximilian Schiffer 53