R -Stack Exchange Project Presentation -API: Accepted Question/Response Time Exploration
Ifeoluwa Kayode, Avinesh Agrawal, Ashutosh Misra
2024-04-23
Team members
Avinesh Agrawal
Ashutosh Misra
Ifeoluwa Kayode
Problem Description
For our project, we have chosen the Stack Exchange API. It has been found that, at times, users experience issues with longer response times.
Our project aim is to utilize data from the Stack Exchange API and analyze question posting times in correspondence to their respected response times.
Original Analytics plan
We aim to utilize machine learning models within H2O AutoML in order
to predict the maximum time users may wait for responses.
Data
will be gathered from diverse endpoints using the Stack Exchange API
from which a data frame will be constructed.
Cleaning,
pre-processing, and EDA of the data will be conducted for the sake of
insights, adhering to tidy principles using tidyverse and
ggplot2.
Modified Analytics plan
Every aspect of the original analytics plan was maintained; however,
rather than using H2O AutoML, it was learned that the process of solely
using H2O would work best for goal.
In order to work with Stack
Exchange API, a app was created in order to obtain a Client ID and Key
for requests to be made.
Key Peer Comments from Proposal
-Collecting information on posting/response times across various
tags.
-Analyzing patterns in question trends and answer acceptance
rates.
-Possibility of potential limitations presented by the use of
AutoML.
Data Summary and EDA Results
Box Plot
Scatterplot: AvgMins vs.TagName
Bar Chart of Software Topics
Sample Data
Description of AI/ML Procedures and Results
Our overall analysis consisted of producing the following predictive models: Decision Tree, Random Forest, and Support Vector Model.
Through a comparative analysis of the performance concerning the accuracy of each model, it was found that the Random Forest model produced the highest level of accuracy among the three with a precision value reaching 0.8.
Visual Comparative Analysis of Model Accuracy
Discussion of Key Takeaways
Through a comparison of topics, it was found that dbt was the tag/topic that received the shortest question/responses times, while Azure was the topic with the longest response times.
The results of this analysis may provide insights into queries regarding why certain topics may engender longer accepted answer/response times than others. It may also pique the question of what kind of tags, questions, and so on are individuals more likely to inquire over.