Nikunj Kotecha

- Developed a LSTM autoencoder model to predict if the user will enter in a state of crisis using Python and Keras
- Explored patterns in user messages with Kmeans and DBSCAN clustering analysis using Python and Scikit-learn
- Generated Wordclouds to identify most frequent words used before and after the state of crisis

- Improved accuracy by 2% with RandomForest classifier for identifying and parsing a handwritten math expression
- Collected the data into a SQL database and performed data cleaning and feature engineering for given expressions
- Implemented classification, segmentation and parsing for handwritten math expressions

- Implemented a depthwise separable CNN neural network to predict breast cancer using Keras
- Generated classification reports such as accuracy, F1 score, specificity, and sensitivity

This project focuses on Segmenting an image consisting of multiple watermelons in it.
It follows the approach of Supervised Learning.

In order to create a model, first provide an image where multiple cut pieces of watermelons are present. Then give few inputs for the skin and foreground of the melons.
The program uses the concept of Mahalanobis distance and segments the melons with respect to skin and foreground and draws a blue line separating the two.

You can use the model… Show more

This project focuses on Segmenting an image consisting of multiple watermelons in it.
It follows the approach of Supervised Learning.

In order to create a model, first provide an image where multiple cut pieces of watermelons are present. Then give few inputs for the skin and foreground of the melons.
The program uses the concept of Mahalanobis distance and segments the melons with respect to skin and foreground and draws a blue line separating the two.

You can use the model generated and pass on any image having melons pieces and the program will use this model and provide a segmented output. Show less

- Developed an algorithm that shows all the positions of a diver in one frame using Background subtraction and OpenCV
- Implemented a background model that could separate the diver and the background in the video

- Developed a CNN based neural network using Tensorflow that recognizes characters in Hindi language
- Visualized, preprocessed and analyzed the given data for Hindi language characters

This project is implemented solely for learning purposes.
The code and project direction is influenced by https://medium.com/@martinpella/customers-tweets-classification-41cdca4e2de

The dataset used for this project is called "Airline sentiment tweets" and can be found here: https://www.kaggle.com/tango911/airline-sentiment-tweets


In this project, first pre-processing steps are taken place. In this text cleaning and any unwanted noise is removed.
The data are split into… Show more

This project is implemented solely for learning purposes.
The code and project direction is influenced by https://medium.com/@martinpella/customers-tweets-classification-41cdca4e2de

The dataset used for this project is called "Airline sentiment tweets" and can be found here: https://www.kaggle.com/tango911/airline-sentiment-tweets


In this project, first pre-processing steps are taken place. In this text cleaning and any unwanted noise is removed.
The data are split into training and test set for better learning of the model.
Text vectorization is implemented after this using the Tf-IFD method and sparse vectors are generated.
Logistic regression is used to train the model.


The model is able to achieve an average of 80% accuracy. It is able to classify tweets into neutral, negative and positive sentiments with this accuracy. Show less