Machine-Learning-Projects

This repository contains Minor Machine Learning Projects. This Projects are completed by me for self learning and hobby purposes.

Contents

• Machine Learning

  • Cat Dog Image Classification: In this project I used keras pretrained InceptionV3 model. I added a Flatten and 2 dense layer for final output. My model got a a validation accuracy of 0.98.

  • Spam Classification using Bert: In this project I created spam classifier using Bert. I used a pretrained model from Huggingface and fine-tuned it. In the end I got Accuracy 1. and f1 score 1.

  • Language Translation using Encoder Decoder: I downloaded the French-english data from http://www.manythings.org/anki/ . After that I preprocessed the data and converted the data into BagofWords but I took characters as features. After that I created Encoder and Decoder layer and saved it into a Model. After training the Model I got an accuracy of 86.3%.

  • Fake-news Classifier: I took the data from kaggle competition and used it to build a fake news classifier model. I used Bidirectional LSTM, MultinomailNB, PassiveAggressiveClassifier, RidgeClassifier and SGDClassifier. And I got a f1 score of 0.9308 using SGDClassifier.

  • Spam Classifier: In this project I created spam classifier using Multinomial Naive Bayes Classifier. I used BagofWords as well as Tf-Idf for vectorising the data. In the end I got Accuracy 0.99 and f1 score 0.97.

  • Next Word Prediction: In this project I took India's data from Wikipedia and trained the model with this data. The model consisted of 2 LSTM, 1 Embedding and 1 Dense layer. It took 5 text sequence and predicted the next word. While training the model obtained a loss of 0.01.

  • PNB stock prediction: In this project I tried to predict stock price of punjab natioal bank(PNB). I visualized the data and tested for trend using linear and polynomial regression. Also tested for stationarity using adfuller. Then to build a predicting model I used RNN with 3 LSTM layers. It took Open price for last 60 days and predicted the price of next day. The MSE obtained after predicting test data was 4.723246.

  • Music Genre Classification: It is a multiclass classification problem where we need to predict music genre using some features. So I precessed the data first by using KNNImouter to deal to null values than I used Labelencoder to convert Categorical data to numerical data. I also removed the outliers from some of the feature and finally scaled the data. I used RandomForestClassifier and LogisticRegressor as the model which gave a accuracy of 61% and 63% respectively.

  • Titanic - Machine Learning from Disaster: Exploratory Analysis of the passengers onboard RMS Titanic using Pandas and Seaborn visualisations. Created two model one using ANN and another using RandomForest. The data was taken from Kaggle. The task was to predict a passenger would survive or not.

  • MNIST - Number Recognition using CNN: The Data is been extracted from Tensorflow dataset. In this project i have used Convolutional neural network to Categorize the Number's image into correct Number Category. This project was made by me while doing a Course on Tensorflow.

  • California_Housing: The Data was provided during one of my course on DataScience. In this Project i need to predict the Sale Price of house using some features. To do so i used 4 different models. Linear Regression, ANN, Decision Tree Regression and Random Forest Regression. Out of this Random Forest performed better than rest.

  • Credit Card Approver: The Data Set has been downloaded from Kaggle. There were 2 data set one Application and credit record which is linked by ID. In this Project i used Random Forest Classifier to predict whether or not to approve Credit Card to a person depending on the details provided.

  • Credit Card Lead Prediction: This code was written during a competition in Analytics Vidya. We need to predict whether a person is interested in credit card or not. I used Deep learning to predict. My ROC was 0.8700320345.