Image Processing and Classification Model using CNN.

Collection of Damage(Scratch Or Dent) car datasets from different different resources.

Labeled the data with Scratch and Dent.

Re-labeled the dataset with which parts is having scratch or dent.

Train the CNN model using that dataset.

Predicting the car has Scratch or Dent and which part is having Scratch/Dent.

In Process- Predict how much cost it will take to repair that part.

Pulling data from Azure SQL database.

Preprocessing on Data.

Filtered last 36 months data and predict the next 6months attritions using Moving Median method and Time Series Model.

Automate the whole process on Azure App Service.

Analyzing the Review and predict the sentiments.

Used Vader Algorithms to predict the 3 types of Sentiments with their percentage value and overall sentence sentiment.

Created corpus of word for Multiple emotions and used that corpus to predict the Emotions w.r.t. Sentence using Almo-Bert model.

Recommending Products Or Services to the Account (Companies).

Applied Recommendation Machine Learning Algorithms to predict the Products which company have more chances to buy in future.

Algorithms used in Recommendation System:- KNN, Collaborative Filtering, Popularity Based Recommendation System.

Also handled the Cold Start problem.

Data collection from Google, preprocessing of data using SAS image action sets, used SAS deep learning action to createour own CNN model and get the accuracy of 96%, also tried multiple pre-trained models developed by SAS.

Trained a custom CNN model using SAS image Actions which helps in recognizing the facial expression of a person, oncethe model is ready we Integrated this with ESP for live input, this will start webcam and start recognizing the faces fromthe live video input using Haar-cascade classifier and that face will pass through the model and start giving the predictionof facial expression of the person on the same live video input with boundary box around the faces and predictedemoticon on top of the box. (Emotion labels – Happy, Sad, Angry, Neutral, Surprised)

Analyzing Air Mauritius Flights reviews data, Applying Text Mining techniques, Used Classification algorithm for classifyingthe reviews as Positive/Negative/Neutral, To improve the accuracy we used Vader algorithm and then Used pre-trainedmodels like BERT for predicting sentiments from the users reviews.

Created a pipeline for All Classification ML Algorithms with Hyper parameter tuning to get the best score from eachmodel. Preprocessing and Model Training will be done just by fitting the data into pipeline.

Processed huge imbalanced data. Leveraged the capabilities of dotData tool. Improved model performance by parameter tuning and optimizing the model. Utilized Neo4j Graph Data Science capability to identify more Gray Market patterns in Data. Created Power BI report and demonstrated it to stakeholders. Built RAG model and integrated with Neo4j Database.

Processed Organization Financial Monthly Data. BU-Level and Region-wise Forecasting Gross Orders. Explored multiple AutoML models available on GCP Cortex and Azure. Evaluated BQML models available on Vertex AI. Analyzed multiple methods, feature engineering techniques, and forecasting model to get concrete model. Leveraged the capabilities of Finance Time Series Forecasting tool to get more accurate predictions/forecasts. Deployed the model on Azure and scheduled monthly run through Control-M-Server.