Description

CSCI4343 – Data Mining Final Project
November 2024
1 Introduction
In this project, you will try to become a coffee data miner. Pour over techniques is a very popular techniques that used in a lot of coffee shop. But to make a delicious coffee is not very simple. For example, to achieve the best coffee brewing performance, different world champion invented different type of strategies. If you like to know, here is a incomplete listed:
• 46 Method
• Rao Method
• Five pour Receipt
• Centre Pour Method
There are many different variations of filters and the design of pour over system, V60, Kalita, Origami, each one has its own features. A delicious cup of coffee often obtained by a very .
With the advance of sensor techniques, now, we could monitor the pour over method via data! We placed a smart scale to measure the water volume increasing throughout the process. You responsibility is: Can we predict the origin of the coffee bean from the data?. You will provide a set of data and the class label
2 Submission Guideline
Please complete the blackboard team member submission via the corresponding blackboard link and including your all team members. Each team can have up to 4 students.
The final project requires you to submit: • a report (up to 3 pages) consists of introduction, methodology you used in the project, and experiment result
• Source code you used and the running result. If you use human crafted feature, please upload the crafted features as a csv file.
• Readme file about how to run the source code
• Feedback of this project
Additionally, you need to mention the responsibility of each student in the project report. Google Colab is highly recommended. For each team we have:
• Team with 1-2 Members: You can choose to do Task 2.1, Task 2.2, and Task 2.5 or Task 2.3, Task 2.4, and Task 2.5.
• Team with 3-4 Members: You need to do all Task 2 tasks.
3 Data Description
Like our Katydid vs. Grasshopper example in the class, we need to analyze the features. A preprocessing step is needed to obtain a set of useful features. You need to conduct the following steps:
An example of the data is shown in the following figure where x-axis is time stamp and y axis is the water volume.

Figure 1: An Example of Brewing Data
The loading code is already provided in the corresponding google colab code. For each time series, it has the following format:
brew data: water volume collected over time. It is stored in an np array: “coffee dataset”.
origin of the coffee bean: For each brew data, we have a label labeling the original of the data. It will be either “Sprout” or “Rwanda”. It is stored in “coffee labels”
4 Task 1: Data Preprocessing
The goal of this project is “build a data mining model such that, given the input of the brewing data, it guess the origin of this cup of coffee. In task 1, you need to complete the following pre-processing tasks:
Task 1.1: Visualize samples (5 pt) Loading data and use python (package matplotlib) to plot the data.
Task 1.2: By visualizing such data, answer which features you think is the best to form a data mining problem? Write down your answer in your submitted report.
Task 1.3: Convert the text label “Sprout” and “Rwanda” to numbers so that we could implement our classification model. To do so, you can form a new label array such that replacing all “Sprout” to 0 and “Rwanda” to 1. This array will be useful for Task 2
5 Task 2: Training Model (15 pt)
After you extracted the feature, next step, we will conduct a human feature vs. machine feature challenge. We have four steps in this part:
Task 2.1: Train a MLP model using all the data labeled data and output is two classes to perform classification.
Task 2.2: Taking cross validation into account, train a MLP model using 60% of labeled data and use the other 40% of labeled data to adjust which learning rate you should use and which epoch you should stop. Write down your observation in your submitted report.
Task 2.3: Train a MLP model to classify the data based on the features you summarized in Task 1.3 and output is two classes to perform classification.
Task 2.4: Similiar to Task 2.2, taking cross validation into account, train a MLP model using 60% of labeled data and use the other 40% of labeled data to adjust which learning rate you should use and which epoch you should stop. Write down your observation in your submitted report.
Task 2.5: I will provide 4 unlabeled coffee data in the later stage of the project. Please write down your model output. bonus 2pt: team who get all 4 samples correct will recieve extra bonus.