Ashish Kumar Email & Phone Number

Noida, Up, In

Noida, Uttar Pradesh, India

Noida Area, India

• Built forecasting models on Depletion(actual sales) data as model was already built for Demand data having more absolute error. Before forecasting, Trend, Variability and Sales Type features were also identified and added to the raw data. In dashboard, all models (XGBoost, CATBoost, ARIMAX, Prophet etc.) were there for client’s better understanding. History correction had also been applied during corona crisis period which varied province to province in Canada• Using AWS S3 bucket… Show more • Built forecasting models on Depletion(actual sales) data as model was already built for Demand data having more absolute error. Before forecasting, Trend, Variability and Sales Type features were also identified and added to the raw data. In dashboard, all models (XGBoost, CATBoost, ARIMAX, Prophet etc.) were there for client’s better understanding. History correction had also been applied during corona crisis period which varied province to province in Canada• Using AWS S3 bucket, downloaded multi-spectral 13 bands, acquired at three different spatial resolutions (10m, 20m & 60m) and resized them at same resolution of 20m for training purpose. Created a CNN network to train the pre-processed data of satellite sentinel L1C tiles (19HDB, 19KBB, 19LYK etc.) of 100x100 Square kilometre. CNN model was trained (on AWS Sagemaker for computational convenience) using Geotiff images and got the Inference on unseen data to find the possibility of a new mine and its geo location• Developed a customised Deep Learning driven model that feeds scientific data(texts) about medicine(s) (documented by medical researchers) and creates customised messages for marketing campaigns with a quality score. It helped in reducing manual intervention by automating the complete process and replacing Linguists. GPU was also used for intensive algorithm training to reduce training time. This custom Deep Learning algorithm was deployed over AWS VM, API gateway. Show less

Noida, Uttar Pradesh, India

• Performed Theme Extraction & Sentiment Analysis on scrapped data of facebook user’s comments for a leading Track racing circuit. Initially training sample of texts(comments) were manually tagged, thereby forming modelling base for training. Leveraged word vectors & TF-IDF (SVM, XGBOOST, Naïve-Bayes, Random Forest)• Created segments of HCP on basis of Decision tree using CHAID to identify segments with growth opportunity. We used attributes of both HCP and patient to understand higher drug… Show more • Performed Theme Extraction & Sentiment Analysis on scrapped data of facebook user’s comments for a leading Track racing circuit. Initially training sample of texts(comments) were manually tagged, thereby forming modelling base for training. Leveraged word vectors & TF-IDF (SVM, XGBOOST, Naïve-Bayes, Random Forest)• Created segments of HCP on basis of Decision tree using CHAID to identify segments with growth opportunity. We used attributes of both HCP and patient to understand higher drug prescription. Show less

Mumbai Area, India

1) Fraud Detection Model:- a) Objective was to identify whether a Health Insurance Claim is a Fraud or not at pre-authorization level b) Detected anomalies in claims and Implemented a logistic Classification Model c) Model is re-calibrated every month & it keeps the best model in terms of Precision/Recall. It’s been also deployed on server which is executed in every 15 minutes and results are shown at Dashboard of Tableau2) Policy Renewal (Persistency… Show more 1) Fraud Detection Model:- a) Objective was to identify whether a Health Insurance Claim is a Fraud or not at pre-authorization level b) Detected anomalies in claims and Implemented a logistic Classification Model c) Model is re-calibrated every month & it keeps the best model in terms of Precision/Recall. It’s been also deployed on server which is executed in every 15 minutes and results are shown at Dashboard of Tableau2) Policy Renewal (Persistency Model):- a) Created a Persistency model(by stacking different classifiers) which puts an ongoing policy into a risk level category(High, Mid and Low) depending upon the customer’s interest in renewing the policy b) A month before expiration of a policy, agent can target only the customers who are less likely to renew Show less

Kolkata Area, India

• Created a Generalised Linear Model on data consisting instantaneous speed, overhead current, voltage, power factor and consumed energy generated by Locomotive Engine of Indian Railways• Decreased Root Mean Square Error value by diagnosing Non-normality, Heteroskedasticity, Non-linearity and Correlated(Time Series) observations in the model• Found optimum speed interval for better energy efficiency by calculating energy consumption per unit distance

Kolkata Area, India

1) Pitched and convinced 78 cab owners to come under the Umbrella of "Namaste Taxi" and trained the existing drivers to use Mobile App2) Analyzed & sorted data of all drivers by Statistical Method and optimized the distribution of cabs within the city by Time Varying Poisson Model3)Online booking facility ​ : Started with a fleet of initial 78 cabs as per their location and level of service increasing their productivity and efficiency

Research Project, Linear Models: An Integrated Approach

B.Tech. And M.Tech. Dual Degree, Ocean Engg. And Naval Architecture/ Engg Entrepreneurship

Activities and Societies: Spring Fest

High School, Mathematics

Education record