Predicting Live Delivery Success along with its Corrective Actions to be taken

Predict early life delays during L2C journey

M2M Central Billing aims to implement a new billing solution to support rating and billing of all M2M customers using Amdocs Partner Relationship Manager (APRM) over the current legacy systems which are manual, inefficient and time consuming and not cost effective. As a Test Analyst, I was responsible for the following: Analyzing requirements and preparing Effort Estimation Preparing test strategy and test plan and monitor the progress against the plan Coordinating with multiple teams like Development Teams, E2E Design, Solution Architects etc. for defects fixes Daily Test Execution Reporting to Customers Client interaction on daily basis for highlights and risk Ensuring the team is aware of the target completion date and give their best to achieve the same Coordinating UAT handover

Predict whether the Pseudo Survey task is required or not during an L2C order journey

Predict the number of defects that are going to be raised for a user story

Self-healing Indicent Management Microbot is visionary AI bot which will serve as human and emulate the process followed by human to fix the incidents and close the tickets all by the Microbot. As a Data Scientist, I am responsible for the following: Brainstorming sessions with E2E Design to understand Domain and current workflow of the manual Incident Management system Analysing the data produced the current system EDA and labelling the observations Modelling using different algorithms like Logistic Regression, K-means and Passive Aggressive Classifier (PA) and found that PA is producing higher accuracy on test data Achieved 80% accuracy on test data so far for the POC we are doing Currently we are incorporating the SOPs to the AI model

P&B Failure Prediction is AI enabled model which will predict Early Life delays during L2C journey which pin points to the failure in plan & build. It will to improve customer experience by providing seamless installation of services and save on reactive infrastructural cost. As a Data Scientist, I am responsible for the following: Brainstorming sessions with E2E Design to understand Domain Studying current L2C Order delays Plotting different graphs to unearth any emerging pattern of delay EDA and Feature engineering Modelling using different algorithms like Linear Regression (for predicting number of days it will get delayed to complete an L2C Order) and Classifiers to be exercised to predict the category of the delay (reason for the delay precisely) Achieved 95% accuracy on training data for Linear Regression model Currently working on RMSE minimization The mature predictive model will predict the category of the delay and number of days it will delay to complete the order

Test Estimation is vital process in BT and to come up with near-real estimate, the defects prediction plays important role to provide estimation which varies less than the benchmark otherwise is 5% of the effort estimated. As a Data Analyst, I was responsible for the following: Analyze the requirement and no. of defects raised per requirement in different release and phases of test life cycle Team up with Code Developers, Test Analysts to understand the complexity of the different requirements Pinpoint the problem area and find the required data for analysis Plot different graphs to build initial hypothesis Exploratory Data Analysis and unfold different insights to support the hypothesis or counter the same and found important features using correlation matrix Build robust predictive model which accurately (98%) identify how many defects will be raised for a particular release successfully using Linear regression model

Analyzed data using Pandas dataframe Plotted data and inferred insights using Matplotlib and Seaborn ROC AUC achieved - 51% Model built using Logistic Regression CV

Analysis done using Pandas, Matplotlib and Seaborn Found important features using Random Forest Modeled using Logistic Regression, KNN, Random Forest and achieved 70%, 99% and 98% accuracy respectively Chosen Random Forest as final model since it was not overfitting and generalized the prediction