Tech enthusiast skilled in Data Science, ML, AI, Web Dev. Certified in Data Analytics & Zero Trust Security.

github

codeforces

devto

Keshab Kumar  • Peerlist

This project predicts the Air Quality Index (AQI) using advanced machine learning techniques with a focus on robust data imputation, outlier handling, and model selection. Multiple Imputation by Chained Equations (MICE) was used to manage missing data with estimators like Ridge Regression, Bayesian Regression, and Elastic Net, ensuring a complete dataset. Outliers were removed prior to imputation to prevent data skew.

Models Implemented:

LinearRegression
Ridge
SGDRegressor
ElasticNet
Lasso
SVR
Decision Tree Regressor
Random Forest Regressor (Top Performer)
Gradient Boosting Regressor

After testing, the Random Forest Regressor yielded the most accurate AQI predictions. Visualizations compared model performance, with Random Forest emerging as the most reliable predictor. This project serves as an effective framework for accurate AQI monitoring.

Air Quality Prediction

In this project, I participated in the Regression of Used Car Prices competition on Kaggle, where the goal was to predict the price of used cars based on various attributes. Leveraging a robust data science pipeline, I conducted exploratory data analysis (EDA) using libraries like numpy, pandas, matplotlib, seaborn, and dython. I applied data preprocessing techniques, including KNNImputer for handling missing values and performed feature engineering to optimize model performance.

For model building, I used LGBMRegressor and XGBRegressor, achieving an RMSE of 72738.3121 with LGBMRegressor and 72704.0094 with XGBRegressor. The final predictions were saved in a submission.csv file. The complete code and analysis are available in my Kaggle notebook:  
https://www.kaggle.com/code/keshabkkumar/car-price-prediction

Car Price Prediction with Machine Learning - Kaggle Playgrou

ClimaSense is an intuitive web application that delivers real-time weather information tailored to the user’s location or any searched city. Utilizing the OpenWeatherMap API, ClimaSense features automatic location detection and manual search capabilities, ensuring users receive the latest weather updates with ease. The application is enhanced by an intelligent chatbot, built using Python frameworks like Streamlit and SpaCy, which responds to user queries about weather conditions. ClimaSense combines simplicity and functionality, making it an essential tool for staying informed about the weather.

ClimaSense

This project, part of a 6-week internship with SkillsBuild and Edunet Foundation (in collaboration with AICTE), focuses on predicting employee burnout using regression techniques. The dataset includes factors like workload, mental fatigue, and WFH status.

Key Steps:

Data Preprocessing: Handled missing values, encoded categorical features, and scaled data.
EDA: Analyzed patterns and correlations.
Model Development: Applied regression models (Linear, Ridge, Lasso) with cross-validation.
Evaluation: Assessed performance using MAE, RMSE, and R².
The model effectively predicts burnout risk, helping organizations enhance employee well-being.

Employee Burnout Analysis and prediction

Imsage is a user-friendly image classification web application designed to simplify and enhance image recognition tasks. Built using Streamlit and TensorFlow, the app integrates two robust machine learning models:

MobileNetV2 (trained on ImageNet): Known for its efficiency and accuracy in image classification.
Custom CIFAR-10 Model: Tailored to classify objects into 10 distinct categories from the CIFAR-10 dataset.
Key Features:

Image Uploads: Users can upload images for real-time classification.
Model Switching: An intuitive interface allows seamless toggling between the two models.
Confidence Scores: Provides predictions with detailed confidence levels for better insights.
Result Export: Option to download prediction results in CSV format for further analysis.
Real-Time Performance: Ensures quick and accurate predictions, making it ideal for both learning and practical applications.

Imsage: Image Classification Web Application

Inscribe

PG-Life is a comprehensive web application developed to streamline and enhance the experience of individuals searching for Paying Guest accommodations. Built on a foundation of seven key web technologies - HTML, CSS, Bootstrap, DBMS, PHP, JavaScript, and React - PG-Life offers a user-friendly and feature-rich platform for users seeking hassle-free PG options.

Key Features:

Responsive Design: Utilizing HTML, CSS, and Bootstrap, PG-Life ensures a seamless experience across devices, from desktops to mobile phones.

Dynamic Database Management: The incorporation of a robust Database Management System (DBMS) ensures efficient data storage, retrieval, and management of PG listings.

Interactive User Interface: Leveraging the power of JavaScript, PG-Life boasts an interactive and engaging user interface, providing a smooth navigation experience.

Server-Side Scripting with PHP: The integration of PHP enables dynamic content generation, facilitating real-time updates and personalized user experiences.

Advanced Functionality with React: The use of React enhances the application's performance by enabling the creation of interactive and dynamic user interfaces.

Explore the codebase and contribute to the development of PG-Life on https://github.com/Keshabkjha/pglife

PG-Life represents a culmination of skills acquired through Internshala Training, showcasing proficiency in a diverse range of web technologies. It stands as a testament to the commitment to creating practical and effective solutions for real-world challenges in the realm of accommodation search.

PG-Life web application

ProfsPortail is a collaborative platform designed for institutes to facilitate seamless communication and management among professors, Head of Departments (HOD), and students. The platform aims to enhance productivity, streamline administrative tasks, and foster a sense of community within departments.