Comprehensive predictive analysis on how Internet Service Prices vary across different geographics, demographics etc. for improved internet access and affordability.
The project can be well understood by this blog
The model is built in an Anaconda Environment (Jupyter Notebook) and Python 3.5.0. All the libraries which need to be downloaded are mentioned in requirements.txt.
- Install Python3 or above using Anaconda or any other method
- Install the requirements using pip install -r requirements.txt in Anaconda Prompt
- Clone this repo to your local machine
- Extract the zip file you downloaded
- Open a Python editor like Jupyter Notebook
- You may create and use a virtual environment to work on this project
- Run CIS545_Project_final.ipynb
The project makes use of 5 csv files downloaded from here, namely:
- speed_price_att.csv - contains internet plans data for AT&T network for major states of USA
- speed_price_att_other_cities.csv - contains internet plans data for AT&T network for Ohio & Tennessee states of USA
- speed_price_centurylinik.csv - contains internet plans data for Centurylink network for major states of USA
- speed_price_earthlink.csv - contains internet plans data for Earthlink network for major states of USA
- speed_price_verizon.csv - contains internet plans data for Verizon network for major states of USA
- Data Pre processing
- Handling Null values
- Handling abnormal negative values
- Date time conversion
- Exploratory Data Analysis & Visualization
- Univariate Analysis & Multivariate Analysis
- Heatmap
- Dynamic plotly geographic maps to understand market share
- Feature Engineering
- Outlier analysis
- Feature Encoding
- Modeling & Hyper parameter optimization
20 different versions of ML and DL models were implemented. The results are given below :
