Projects

Valuation Envisioner   image

    Machine Learning, Regression, Hyper-Parameter Tuning
  • Purpose is to design end-to-end predictive model that can accurately anticipate selling price of a car based on its features in order to make informed purchase decision.
  • Plotted powerful multivariate visualization using matplotlib and seaborn which refine insights of selection that uncovered positive and negative correlation.
  • Raised predictions with ensemble Random Forest regression method and as a result of RandomizedSearchCV using 3fold Cross-Validation improved efficiency of 5% that received the best score by reducing computation time.
  • Built free, scale-ready and just a click away web app on user-friendly Paas infrastructure.

                         

Bust and Boom Perk   image

    Machine Learning, Exploratory Data Analysis
  • Goal is to predict GDP per capita for each country.
  • Cleaned & Pre-processed data using various statistical techniques and performed EDA which led to significant insights of countries with highest GDP as Western Europe through Plotting of Heatmap, Pairplot and Jointplot Visualization that enlighten key factors as migration and literacy.
  • Conducted splitting, evaluation and optimization to compare which of the four regression algorithms were at best in prediction.
  • Secured Finest RMSE Score for Random Forest model and deployed on Streamlit as well as Heroku Cloud Platforms.

                     

Customized Airfare API   image

    Machine Learning, Predictive Analysis
  • Aim is to estimate fare of the flight to help traveler to decide as per convenience.
  • Implemented EDA techniques to restructure columns to appropriate datatypes using extraction, merging, onehot encoding and creating dummies which gave rise to most important feature as total stops that act as stepping stone for high prediction score.
  • Tuned parameter which improved RF regression model score by 70% using most suitable metric.
  • Drove right from preprocess till deployment of a practically handy one-click in interest of end-users.

                     

S2 - Sports Scrutinizer   image

    Machine Learning, Pickling
  • Objective is to predict final score in a limited overs cricket match.
  • Procured 8% greater data integrity that directly impacted success rate of project by scaling down redundancy.
  • Built, trained and tested reliable regression model followed by storing as pickle file that directed noteworthy fastidious,resourceful.
  • Productionized scalable application interface on Heroku cloud.

                     

Ms.Salary Expert   image

    Machine Learning, Scraping, Pipeline
  • Intent is to predict salary of given job role when given specifications that reveals feasibility.
  • Collected and created own dataset from unstructured data through scraping data using selenium.
  • Combination of cleaning and visualization fostered location as striking factor.
  • Scaled down 25% manual errors through automated ML which used customized pipeline that sparked in time and resource saving app.

                 

Wealth Vow   image

    Machine Learning,Classification, Multiple Deployments
  • Goal is to detect forged bank note accurately to escalate user experience.
  • Statistical algorithm of Random Forest Classification resulted in obtaining remarkable accuracy of 99%.
  • Productionized multiple deployment interface on platform such as Flask, Flasgger, Streamlit and Heroku.

                     

GroBiz   image

    Machine Learning, Self-evident App
  • Goal is predict sales amount that help to stock up in order to fulfill demands.
  • Plotted statistical envisage on just click-away interface using prominent libraries matplotlib and seaborn that strengthen decision making process even for non-tech person. Also, machinate engaging icebreaker word cloud.
  • Converted varying scaled data using standardization that speeded up dissemination across state-of-art leading to amplified model’s accomplishments.
  • Hosted an app by creating public URL on free and young Streamlit Cloud.

                     

Smart Farming   image

    Deep Learning, CNN
  • Goal is to detect plant and leaf is diseased or fresh and provide recommendations.
  • Segmentation through thresholding method helped to reduce complexity and able to extract object of interest needed for model building.
  • Produced Data Augmentation followed by Customized CNN for training images even faster using GPU.
  • Able to pull off 98% accuracy by precisely classifying to help farming community to boost crop productivity, in making effective decisions and efficiently to protect their crops from heavy loss due to a vast spreading of disease.

                     

AIBasketball Detector   image

    Deep Learning, OpenPose, Faster R-CNN
  • Goal is to detect object for basketball shot and pose analysis.
  • Designed using OpenCV and flexible Tensorflow with support of GPU as well as CPU Computation resulting in scalable and speed up process.
  • Trained a model with Faster RCNN as the state of art approach of OpenPose to detect human coordinates which calculated angel through keypoints.
  • Established an app by deploying on Heroku.

                        

Hall-mark Un-masker  

image

    Deep Learning, Optimized Curve
  • Goal is to recognize brand of the luxury cars.
  • Detected high-level features using pre-trained models in Keras.
  • Built CNN model which resolved problem of training extremely deep neural network making it faster.
  • Improved under-fitted model accuracy 3times more with the help of epochs that tuned into optimized curve.

                     

Superfluous Radar   image

    Machine Learning, Deep Learning, RF-XGBoost-DTree, OpenCV-Tensorflow, Classification
  • Aim is to detect type of waste in real-time.
  • Improved 2times Evaluation Score by comparing 3 tuned models.
  • Constructed live waste identifier with glaring OpenCV features as Data Augmentation, Transformation and Calculated Histograms.
  • Not only techinical learning but also escalated insights on revolutionary gadgets for visionary society.

                 

Live Video Sketcher   image

    Machine Learning, Computer Vision, Image Processing
  • Built live video sketcher application using prominent OpenCV, Keras library with Gaussian Blur functionality led to interactive learning.

                 

Monica Desai