Machine Learning Projects

What this Repository is About

This Repository contains all of the Machine Learning as well as Deep Learning Projects, that i have made for any Competition or have done only for Practice. In this repo. all the projects will have at least one Complete Notebook, and the number of inComplete notebooks, i don't know. In some of the Projects the Dataset is not uploaded, along with the notebook, it is because the Dataset is very huge in size, but I have mentioned the Source or link of the dataset, if by mistake i have not mentioned it in any of the projects, let me know!. Hope you get benefitted with this repo of mine.

List of Projects Till Date

1. HackerEarth: Emotion Detection - Tom and Jerry Cartoon

• Detects emotion of Tom and Jerry characters in the video, using CNN with 25,696,261 parameters. Dataset used is from HackerEarth Competition with name "Detect emotions of your favorite toons".

• Training Accuracy: 0.9904

• Validation Accuracy: 0.6556

• HackerEarth Score: 26.98841 ( without Image Augmentation )

2. HackerEarth: Mother's Day with Machine Learning

• Extracts sentiment hidden in the tweets posted during Mother's Day, using NLTK, tf-idf, SGD Classifier. Dataset used is from HackerEarth Competition with name "Machine Learning with Mother's Day".

• Training F1 Score: 0.7489

• HackerEarth Score: 0.3680

3. HackerEarth: Identify the Dance Form

• Detects the Dance form in which the person in the image is posing, using VGG-16 pre-trained model, by transfer learning with 21,139,528 parameters. Dataset used is from HackerEarth Competition with name "Identify the Dance Form".

• Training Accuracy: 0.9841

• Validation Accuracy: 0.9075

• HackerRarth Score: 57.51433 ( with Image Augmentation )

• Training Accuracy: 0.9940

• Validation Accuracy: 0.9144

• HackerRarth Score: 62.61918 ( with Image Augmentation and K-Fold Validation )

4. Image Classification on CIFAR10

• Detects objects in the frame, from one of the ten classes included in the training dataset i.e. [Airplane, Automobile, Bird, Cat, Deer, Dog, Frog, Horse, Ship, Truck] , using CNN with 1,250,858 parameters. Dataset used is CIFAR-10.

• Training Accuracy: 0.7256

• Validation Accuracy: 0.7332 ( without Image Augmentation )

5. Kaggle: Handwritten Digit Recogniser MNIST

• Detects the Handwritten digits in the given image, using CNN with 600,810 parameters. Dataset used for training is MNIST.

• Validation Accuracy: 0.98904

• Kaggle Score: 0.98878

6. Kaggle: Titanic - Predict the Survivor