Skip to content

An application that can correctly classify four main types of public transportation in Lagos state, Nigeria.``

Notifications You must be signed in to change notification settings

itsekiri-tatashe/lagos-transportation-classification

Repository files navigation

Lagos Public Transportation Classification witth AI


The aim of this project is to build an application that can correctly classify four main types of public transportation in Lagos state, Nigeria. These vehicles include

Check it Out!

  • Bus Rapid Transit (BRT)

    • BRT Bus

  • Danfo / Bus

    • Danfo Bus

  • Keke Napep / Keke Maruwa / Keke Tricycle

    • Keke Napep

  • Okada Bike / Motorcycle

    • Okada Bike

The data was scraped using the Google Image Scraper tool to extract all the images from Google


  • Extracted the data

  • Arranged and label the data

  • Performed various Data Preprocessing techniques such as converting images into tensors and feature normalization/scaling

  • Used Convolutional Neural Networks (CNN), Data Augmentation and Transfer Learning / Pretrained Model to achieved best performance

  • Tuned MobileNet V2 achieved best performance with 97% traing accuracy and 95.6% test accuracy


Model Performamce

Accuracy with Confusion Matrix was used to evaluate performance. Data Augmentated CNN performed worse than Base CNN and Pre-trained Model

  • Base CNN
    • Training Accuracy : 95.7%
    • Testing Accuracy : 69.7%
    • Confusion Matrix CF of base

  • Data Augmentation + Base CNN
    • Training Accuracy : 60%
    • Testing Accuracy : 66.9%
    • Confusion Matrix CF of data aug

  • Tuned MobileNet V2
    • Training Accuracy : 99%
    • Testing Accuracy : 93.5%
    • Confusion Matrix CF of base

Model Deployment

The final model with the best score was deployed on a web application built with Streamlit

Web application of the model


About

An application that can correctly classify four main types of public transportation in Lagos state, Nigeria.``

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published