Intro to Predictive Analytics Using Python - University of Pennsylvania

 

Intro to Predictive Analytics Using Python

 - a course by the University of Pennsylvania 

My dairy about the journey of studies:

• Instructor: Brandon Krakowsky
• Coursera Course - Intro to Predictive Analytics Using Python
• 
  

 

Intro to Predictive Analytics Using Python - 3 Modules

• There are 3 modules in this course

  "Introduction to Predictive Analytics and Advanced Predictive Analytics Using Python" is specially designed to enhance your skills in building, refining, and implementing predictive models using Python. This course serves as a comprehensive introduction to predictive analytics, beginning with the fundamentals of linear and logistic regression. These models are the cornerstone of predictive analytics, enabling you to forecast future events by learning from historical data. We cover a bit of the theory behind these models, but in particular, their application in real-world scenarios​ and the process of evaluating their performance​ to ensure accuracy and reliability.​ As the course progresses, we delve deeper​ into the realm of machine learning​ with a focus on decision trees and random forests.​ These techniques represent a more advanced aspect​ of supervised learning, offering powerful tools​ for both classification and regression tasks.​ Through practical examples and hands-on exercises,​ you'll learn how to build these models,​ understand their intricacies, and apply them​ to complex datasets to identify patterns​ and make predictions. Additionally, we introduce the concepts​ of unsupervised learning and clustering, broadening your analytics toolkit,​ and providing you with the skills to tackle data without predefined labels or categories.​ By the end of this course, you'll not only have a thorough understanding​ of various predictive analytics techniques,​ but also be capable of applying these techniques to solve real-world problems,​ setting the stage for continued growth​ and exploration in the field of data analytics.

• 
  
  
•                

  
  
  
• 

  
  
•                

  
  
  
• 

  
  
•                

  
  ***** -------------------------------------------------------------------------------------------------------- *****

• Module 1: Introduction to Predictive Analytics and Regression

(Click the picture to enlarge)

• 03-Oct-2025: Module 1: Intro to Predictive Analytics & Regression - How to use Data - Specialization Intro
• 
  
• 
  
•  
• MOCC1: 
  
•  
• MOOC2: 

 

• MOOC3: 

• 03-Oct-2025: Module 1: Intro to Predictive Analytics Using Python
• 
  
  
  
• 

  
  
  
• About the instructor: 

  

  
  
• Supervised Machine-Learning: 

  
  
  
• 

  
  
  
• 

   
  

  
  
  
• 

  
  
  
• Oct-10-2025: Module 1 - Lesson 2: Supervised Predictive Models
• Typical Machine Learning Pipeline: 
  
  
  
• 

  
  
  
• 

  
  

  
  
• Oct-13-2025: Module 1 - Week 1 - Linear Regression
• 
  
  
  
• 

  
  
  
• 

  
  
  
• 

  
  
  
• 

  
  
  
• Script: >> Recall the ML pipeline that ultimately produces a model f. In Linear Regression, we make an assumption that the model is a linear model. In other words, linear regression assumes a linear relation between some given inputs and the target output. For example, imagine we want to predict the age of a customer based on the number of orders they've made for a certain product. In this case, the number of orders they've made is the input variable and the age of the customer is the output variable. To perform linear regression, we gather data on various customers, including their orders and ages. We plot this data on a graph with the x-axis representing the number of orders and the y-axis representing the age. 
  Each data point is represented by a blue circle on the graph. The goal of linear regression is to find a line that best fits the data points. This line is represented by a red line on the graph. The position and slope of this line are determined through a mathematical process that minimizes the distance between the line and the data points. Once we have this line, we can use it to make predictions. For example, if we have the number of orders for a new customer, we can use the line to estimate their age. The line represents the learned model that allows us to predict the output variable age based on the input variable number of orders. 
  In linear regression, the model parameters refer to the values that determine the specific characteristics of the linear model. These parameters define the slope and intercept of the line that represents the relationship between the input variable and the output variable. Let's go back to the example of predicting customer ages based on their number of orders. In this case, the parameters of the linear regression model are as follows. The slope represents the change in the output variable age for a unit change in the input variable number of orders. The intercept is the point where the line intersects the y-axis. How do we find the most optimal model parameters? 
  If the training dataset has a significant trend, we can easily draw out a line that looks optimal rather quickly. But for noisy real-life datasets, finding the most optimal model is not as straightforward. During the training process of linear regression, the algorithm learns the optimal values for these parameters by minimizing the difference between the predicted values of the model and the actual values from the training data. We need a metric that measures how well the model performs. For machine learning problems, a loss function is usually defined to evaluate the model. Intuitively, the loss function would have small values if the predicted output is close to the desired output and large values otherwise. A commonly used loss function is the Mean Squared Error or MSE Loss Function. 
  This loss represents the average squared difference between the predicted output and the desired output across all samples in the training dataset. The optimal linear regression model minimizes the MSE loss. In practice, we can use the mean squared error function in the scikit-learn machine learning library to calculate the MSE. A simple linear regression involves one independent variable and one dependent variable. A multivariate linear regression extends to include multiple independent variables. Overfitting occurs when the learned model fits the training dataset very well but fails to generalize to new examples. The model on the right has no loss on the training data but does not actually capture the patterns of the dataset. 
  Overfitting is a common issue that can occur when using complex models. When a model becomes too complex, it may start to capture noise or random fluctuations in the training data, leading to poor generalization to new unseen data. To detect overfitting, a training test split protocol is typically used where we use holdout test data to estimate loss on new unseen data. Here's what the process looks like. Step one, randomly shuffle the data set and split the data set into a training set and testing set. Step two, train the model on the training set. Step three, evaluate the learn model on both the training set and the testing set to obtain a training loss and a test loss, also called generalization loss. 
  Other metrics can also be used. A good model should have both low training loss, which indicates that it fits well to training data, and low test loss, which indicates that it generalizes well to unseen data. 
• The Mean Square Error (MSE) Loss Function: 

  
  
  
• In practice, the function mean_squared_error can be found in scikt-learn: 

  
  
  
• Two Types of Linear Regression: 

  
  
  
• Overfitting & How to assess it: 

  

  
  
  

  
  
• Overfitting & How to assess it:  

  

  
  
• Types of Linear Regression

  • Simple Linear Regression: Involves one independent variable and one dependent variable.

  • Multivariate Linear Regression: Extends to include multiple independent variables.
• Polynomial Linear Regression: Allows for non-linear relationships between the dependent and independent variables by incorporating polynomial terms, enabling the model to capture non-linear patterns in the data
• 

  
  
  
• Regularized Linear Regression: A form of linear regression that addresses multicollinearity (high correlation between independent variables) and helps mitigate overfitting by adding a penalty term to the loss function that controls the complexity of the model.

  Some common regularized linear regression and corresponding loss functions are:

  • Ridge Regression: 

  

  
  

  • Lasso Regression:   

  

  
  

  • Elastic Net Regression:  

  

  
  

  where w represents the model parameters and r, 𝜆 are hyperparameters that determine the extent of penalization.
• Oct-16-2025: 💻 Coding Demo: Loading the Data and Exploring the Data 💻
• My Github code demo Downloads (please fork it to run in your Codespace in Github) (if you want to run in Github, to prevent from overwriting the original codes).
• In this coding example, we'll walk through applying and analyzing a linear regression model on the California Housing dataset using the scikit-learn library. Scikit-learn, also known as SKLearn, is a popular open source machine learning library in Python that provides a wide range of tools and algorithms for various machine learning tasks. Before we start, let's first load up our dataset. Instead of using an external source, we'll directly fetch the dataset from SKLearn using the method, fetch_california_housing. When we execute this method, we'll give a parameter as frame, and we'll set this to true to indicate that we're loading the dataset into the format of a Panda's dataframe. We'll fetch that dataset and then we'll store it in our variable called california_housing. Once the dataset is loaded, we'll first take a look at the available description of the dataset. 

  To do that, we can access the DES CR attribute. The dataset contains aggregated data about housing values in each district in California. There are a total of over 20,000 samples where each sample has eight numeric attributes that can be used as predictors and one target, which is also numeric. That represents the median house value of a particular district in California. Let's have a closer look at the dataframe to understand what the detailed values of each of these attributes look like, as well as what the target will be. We'll load our dataframe from the dataset that we fetched. We can do that by accessing california_housing.frame. 

  We'll look at the first five rows in our dataframe. Take the dataframe, housing_df.head, and then we'll see the first five rows. Here we can see our eight attributes. We also have our target that we want to predict called median house value, and all the attributes are numeric values as shown in the Pandas dataframe. Now that we've had an initial look at the data, let's formulate the business problem we'd like to solve using the data. Given the outlined features, we would like to build a supervised machine learning model that predicts the median house value for a given block group using regression. As you recall, linear regression is a supervised machine learning model, and we're using a supervised model because we have labeled data. 

  We know what the median house value is for each sample to train the model, and we're using a regression because we have numeric continuous values as our target rather than categorical values. Prior to training the model, it always helps to explore the data a bit more to deal with any unwanted values and verify that applying a linear regression model is indeed reasonable for this scenario. Let's first have a look at the data using the info method. We'll apply that to the housing_df dataframe. This shows us what the data types are for each of the attributes in the data, as well as a count of the non null values for each of the attributes. Null values correspond with potential missing values in the dataset. Upon observation, we'll notice that we're really lucky. 

  This dataset only contains numerical features, so they're all encoded as floats, and there are no missing or null values in the dataset. Let's have a look at the statistics for the numeric features of the dataset. We can do this by applying the describe method to the dataframe. We'll take the dataframe, we'll call describe. This will show a number of statistics for each of the attributes in the dataset. We have the count, mean, standard deviation, min, max, and various quantiles. Using these statistics, we can easily get an understanding of what the distributions for each feature look like, and how the different distributions relate to each other in terms of scale. 

  Another good way to explore this is to plot a histogram for each of the numeric features to compare the distributions between each feature. A histogram shows you the number of instances which is represented by the vertical axis across a given value range, which is represented by the horizontal axis. A dataframe actually has its own built in histogram method. We'll take the dataframe, we'll call hist. We'll give a bins value of 50, and we'll give you the figure a size of 12 by eight. So we could see everything clearly. Here we have our histograms. 

  We'll zoom out a little bit so we can see everything in place. Let's adjust this. We have histograms for each of the features as well as our target variable here. From these plots, we'll notice that the features have varying scales, meaning that some features have very large ranges 0-15,000. Others might have ranges with smaller values 0-3,000. This could cause the learned model to be biased towards a certain feature over others. To prevent this, feature scaling is a very helpful pre-processing method that can be applied. 

  Another observation is some of the features are tail heavy, indicating that there are values that extend to the right of the median more than to the left. This could make it harder for the model to detect patterns in the data. Finally, for median house value and housing median age, you can notice that the values are being capped, meaning that once they exceed a certain value, so here it would be five, the values are capped at five. Depending on the problem that you're working with, you might or might not want to see your model learn this capping behavior. Let's zoom back in. The insights above can give you some guidance on what features you want to use in your final model and what data processing methods you might need to use. 
• Scikit-learn is a popular open-source machine learning library in Python that provides a wide range of tools and algorithms for various machine learning tasks.
• Running error: new version Python 3.13.x:  

  

  
  
• How to solve? Run on earlier versions. Here are what some idea from AI chatbots: 

  

  
  
• 

  
  
  
• 

  
  
  
• 

  
  
  
• My solution / my point of view: the compatibility issues of the new version of Python 3.13.3: 

  

  
  
  
  
• 
  

• Module 2 - Decision Trees and Introduction to Advanced Predictive Analytics and Random Forests
• 
  
  
  
• 

  
  
  
• Module 3:- Introduction to unsupervised learning and clustering
• 

  
  
  
• 

  
  
  
• 03-Oct-2025: Module 1: Intro to Predictive Analytics & Regression - How to use Data - Specialization Intro
• 
  
  
• 

  
  
• MOOC1: 

  
  
• MOOC2: 

  
  
  
• MOOC3: 

  
  
• =================================================================

• Others:

• 14-Oct-2025: YouTube 木子AI研究所: 哪個AI做PPT最強？結果震驚了我！Skywork/Manus/Kimi/Gamma PPT 能力測評
• 

  
  
  
• 

  
  
  
• Prompt: 我是一名高中語文老師,我需要教學《赤壁賦》, 現在需要製作課程ppt文件,附件為我的教案,請你根據這些信息,幫我生成一套ppt   <測試模式說明 >
• 教案 also done by AI, before the test.
• Skywork:  全球唯一能做 Deep Research (深度研究) Slides的AI ppt 工具
• 

  
  
  
• 

  
  
  
• Source of Info (shown on the right):  

  

  
  
• Other Special Feature: Upload your own PPT template: 

  

  
  
• Not Free - Price Table: Basic USD 19.9  

  

  
  
• 

  
  
  
• 

  
  
  
• Manus: (04:06): 

  

  
  
• 

  
  
  
• 

  
  
  
• Kimi: (05:09): Totally free.: 

  

  
  
• 

  
  
  
• Gamma: (06:20) - Poor and not free. USD 20 per month: 

  

  
  
• Conclusion for above four AI agents: