Skip to content

Latest commit

 

History

History
87 lines (56 loc) · 5.64 KB

README.md

File metadata and controls

87 lines (56 loc) · 5.64 KB

Student Adaptivity Prediction Model

Repository Overview

  • AdaptivityClassifier.sav: Saved model trained by notebook and used for API
  • AdaptivityClassifierTraining.ipynb: Notebook for training the model and visualizing the data (reccommended to run in Colab)
  • ExampleQuery.json: Example .json file, this body can be provided to the API endpoint to produce a prediction
  • ModelServing.py: Code that sets up the API endpoint
  • README.md: Explanation of problem overview, dataset, training and deployement
  • data.csv: Data we use for training the model
  • requirements.txt: Run in clean Python 3.9 environment to install necessary packages for running the API

Problem Description and Overview

This repository seeks to create and serve a model that infers a student's adaptability based on the student's socio-demographic factors. The dataset is a table of the socio-demographic factors of many foreign-born student and their respective ability to adapt to an IT-centric education.

Dataset Details

The dataset contains the data of 1204 students, each with 13 socio-economic values and a 14th "adaptivity" level based on those aformentioned values. The 13 values themselves are:

  1. Gender: Boy or Girl
  2. Age: 6 different 5 year age ranges spanning from 6-10 years old to 26-30 years old
  3. Education Level: School (yet to reach Higher Education), College or University
  4. Institution Type: Government or Non-Government
  5. IT Student: Yes or No
  6. Location: Yes or No
  7. Load-Shedding (inconsistent electricity supplied due to surrounding economic factors): Yes or No
  8. Financial Condition: Poor, Mid or Rich
  9. Internet Type: Mobile Data or Wifi
  10. Network Type: 2G, 3G or 4G
  11. Class Duration: 0 hours, 1-3 hours or 3-6 hours
  12. Self LMS (Whether they use self-teaching modules): Yes or No
  13. Device: Computer, Mobile or Tablet

Additionally, each student is given a 14th "Adaptivity" value, that can be either "Low", "Mid", or "High". We seek to predict the Adaptivity value of students based on the above 13 values provided.

Analyzing the data, choosing the classifier, and training the model

From a simple data analysis (further details and visualization can be found in the colab provided by .ipynb file in this repository), there are no particularly strong correlations between any two features in the dataset. Furthermore, some features are skewed more heavily towards one particular value (but no specific feature is greater than 5 to 1). However, there are no gaps/null values in the dataset and every student has all 14 values accounted for.

All values contained in the dataset are non-numeric, so during the model training and when the saved model is queried, we transform the data from non-numeric to corresponding numerical values.

The dataset was divided with a simple 70/30 split, and trained using 10 different classifiers (Logistic Regression, Linear Discriminate Analysis, Decision Tree, Random Forest, Bagging, Gradient Boosting, ADA boost, XGB, Extra Trees, and a Voting Classifier composed of Extra Trees/Random Forest).

Comparing the classification report, the Voting classifier composed of the Extra Trees and Random Forest classifiers returned the highest scores (93% Train Accuracy and 91% Test Accuracy), and thus was chosen as the model we would use for serving.

How to install dependencies

For deploying the API from a clean Python 3.9 environment, this repository contains a requirements.txt that can be installed with the following command line command (from the directory containing the requirements.txt):

$ pip install -r requirements.txt

With that we should be able to serve the API using the files from this repository. An example .json file is provided that models the type of query this model accepts.

How to train the model, run the serving code

A saved version of the model is provided already to be used by the API. If you wish to generate a file on your own, or modify that way we train the model:

1. Open the .ipynb in colab

Screenshot_7

2. Select "run all" from "runtime" in the toolbar

Screenshot_3

3. Download the data.csv from this repository and provide it in the first cell of the notebook

Screenshot_4

4. After the above steps, a .sav file should be produced in the notebooks directory which you can download to your local directory

Screenshot_5

Once you have the .sav file of the model, run the API python file from the same directory, and that should bring up the API endpoint

How to query the model

With the API endpoint up and running, we can run single sample queries against the model and return it's prediction. The script accepts json inputs that match the format of the data in the dataset. Query data is case-sensitive and requires exact spelling, so confirm with the of the csv file when an error arise (the api is written to return an error when there is an issue with a given input field).

We use the following for the API Endpoint: http://127.0.0.1:5000/JsonQuery

Post Requests can be made either from the command line or any other software for making API calls (see example of PostMan below)

Screenshot_6