Arash Mahmoudian

Data Scientist

4years in AI/ML

6/10Data Wrangling & Cleaning

5years in tech

17years of work

Overview

A data-centric practitioner at a middle level specializing in end-to-end classical ML and time-series predictive analytics. The strongest proven skill is practical feature engineering and applied evaluation as shown by the accident severity notebook (datetime/feature extraction, dummy encoding, correlation pruning) and the time-series notebook (SARIMAX fitting, rolling forecast and RMSE/P-RMSE evaluation). There is little evidence of production MLOps, deployment, experiment tracking, systematic hyperparameter search, or novel research contributions.

Phone

Timeline

WiFi Data Engineer & Network Analytics · Middle

Chater Communications · Full-Time

Feb 2024 to Present 2 Years 9 Months Denver In office

Built Python-based analytics tools to monitor WiFi network performance across large customer deployments. Created data pipelines that merge multiple telemetry sources for WiFi performance analysis. Developed visualization tooling for WiFi topology, mesh networking, and connected-device analytics, supporting KPI tracking and network performance investigations.

Python

Wi-Fi

Bellevue University

Master's Degree · Data Science

2022–2023 Bellevue, Nebraska

Senior RF Data Engineer · Senior

Dish Wireless · Full-Time

Mar 2022 to Nov 2023 1 Year 8 Months Denver In office

Developed an O-RAN observability platform to monitor network KPIs such as call drops, handovers, RSRP, SINR, and geolocation-based performance metrics. Designed ETL workflows to ingest and transform configuration data for nationwide network audits and built Python tools for configuration validation and automation. Implemented anomaly-detection and performance analysis machine learning models and built fault-monitoring solutions using Kafka event streams and AWS S3.

ETL/ELT

Python

Anomaly Detection

Machine Learning

Apache Kafka

Amazon S3

5G NR RF Engineer III · Middle

Samsung/Verizon · Full-Time

Jun 2021 to Mar 2022 9 Months Columbus In office

Created dashboards and automation utilities to monitor 5G network performance using production databases. Developed Python applications to process drive test reports and validate RF performance metrics. Supported nationwide 5G network deployment and migration activities.

Python

Cellular Network Engineer · Middle

TopCom, RighTel, MTN & NAK · Full-Time

Sep 2009 to Sep 2019 10 Years Tehran In office

Performed RF analysis and parameter optimization to improve 2G, 3G, and 4G mobile network performance. Supported KPI improvement through performance analysis and feature implementation across nationwide optimization efforts for large-scale networks. Collaborated on network optimization projects and ongoing operational improvements.

Technical skills

Languages

Python

C++

SQL

DevOps

Amazon S3

AWS

Wi-Fi

AI/ML

Machine Learning

Scikit-learn

Pandas

NumPy

Statsmodels

Anomaly Detection

Analytics

ETL/ELT

Matplotlib

Seaborn

Power BI

Other

Apache Kafka

MySQL

GitHub

Git

Rest API

Jupyter Notebook

Tableau

Classic ML

Time Series

Middle AI/ML Engineer Confidence: Medium Data-centric

Model Architecture & Training

3/10

How well models are designed and trained

Classical model building and evaluation are present (decision trees, random forest, SARIMAX) but no custom training loops, advanced hyperparameter search, or novel architectures.

Evidence

Data-Mining-Accident-Severity/Data Mining_Accident_Severity.ipynb: DecisionTreeClassifier and RandomForestClassifier training and evaluation (accuracy_score, confusion_matrix sections)

Time-Series-Predictive-Analytics/Time-SeriesPredictive Analytics.ipynb: SARIMAX model fitting, auto_arima usage and forecast code (model = SARIMAX(...); model.fit())

Data Pipeline & Feature Engineering

4/10

How data is prepared for models

Substantial feature engineering and large-scale data handling including datetime processing, derived features, categorical encoding, null-handling and correlation-based column pruning.

Evidence

Data-Mining-Accident-Severity/Data Mining_Accident_Severity.ipynb: Start_Time/End_Time conversion, Duration calculation, street type extraction and df['Street2'] creation

Data-Mining-Accident-Severity/Data Mining_Accident_Severity.ipynb: get_dummies on categorical features, drop columns with >40% nulls, df.fillna(method='ffill') and correlation-based to_drop computation

Experimentation & Evaluation

3/10

How results are measured and tested

Clear evaluation practice with train/test splits, accuracy/sensitivity/confusion-matrix for classification and RMSE/P-RMSE for forecasts, but no experiment tracking, cross-validation or systematic hyperparameter tuning.

Evidence

Data-Mining-Accident-Severity/Data Mining_Accident_Severity.ipynb: train_test_split usage and confusion matrix display with sensitivity calculation

Time-Series-Predictive-Analytics/Time-SeriesPredictive Analytics.ipynb: RMSE and P-RMSE calculations for simple and rolling SARIMA forecasts

MLOps & Deployment

1/10

How models are shipped to production

Almost no MLOps or deployment artifacts; only high-level mentions of deployment and error handling without serving, serialization, or CI/CD code.

Evidence

Time-Series-Predictive-Analytics/Time-SeriesPredictive Analytics.ipynb: 'Learning objectives' section lists 'Deployment: Error handling' but there is no implementation of model serialization or serving

Computational Efficiency

3/10

How efficiently computing resources are used

Awareness and mitigation of compute/memory constraints is evident (subsetting, aggregation, weekly grouping, rolling retrain), but no advanced efficiency engineering like batching, quantization or GPU optimization.

Evidence

Data-Mining-Accident-Severity/Data Mining_Accident_Severity.ipynb: selection of Bing sub-dataset to reduce CPU/memory footprint

Time-Series-Predictive-Analytics/Time-SeriesPredictive Analytics.ipynb: comments about MemoryError, use of weekly aggregation and rolling forecast to trade compute vs. accuracy

Research Depth & Innovation

1/10

Depth of research and new ideas

No original research depth or novel algorithmic contributions; uses standard statistical/time-series methods and libraries with practical adjustments.

Evidence

Time-Series-Predictive-Analytics/Time-SeriesPredictive Analytics.ipynb: SARIMA modeling and use of auto_arima with rolling forecasts (standard methods rather than custom algorithms)

Industries

Government· Middle

Transportation & Logistics· Middle

Technologies

Classic ML

Python· since 2021 · Senior

SQL· since 2026 · Junior

C++

MySQL· since 2026

Git

AWS· since 2026

Anomaly Detection· since 2022

GitHub

Amazon S3· since 2022

Wi-Fi· since 2024

Machine Learning· since 2022

Recommendations

Develop production-ready pipelines: add model serialization, reproducible training scripts, and lightweight serving (REST endpoint or batch inference) for the best-performing models.
Introduce experiment tracking and systematic tuning: add W&B/MLflow logging and cross-validation or grid/search-based hyperparameter search for classifiers and SARIMA variants.
Harden data engineering: add unit tests for preprocessing steps, handle edge cases more explicitly instead of blanket ffill, and build data validation checks (Great Expectations or similar).
Optimize compute for forecasting: experiment with aggregated features, faster forecasting libraries or incremental learning approaches to replace repeated full retraining for rolling forecasts.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

arash-mahmoudian Data-Mining-Accident-Severity Time-Series-Predictive-Analytics

Middle Data Scientist Confidence: Medium ML Practitioner

Applied ML practitioner at a middle level specializing in practical predictive modeling and time-series forecasting. The strongest proven skill is applied time-series modeling and rolling-forecast implementation as shown in Time-SeriesPredictive Analytics/Time-SeriesPredictive Analytics.ipynb. The main limitations are notebook-first engineering with limited reproducibility, sparse hyperparameter tuning and calibration, and basic imputation strategies that are not production-hardened.

Statistical Rigor

5/10

Correct use of statistics

Basic to moderate statistical rigor: stationarity tests, ACF/PACF inspection and residual analysis are present for time-series work and sensitivity/ confusion-matrix metrics are computed for classification, but there is limited uncertainty quantification, no formal hypothesis testing workflow beyond ADF, and no adjustments for multiple comparisons or confidence intervals for key estimates.

Evidence

Time-Series-Predictive-Analytics/Time-SeriesPredictive Analytics.ipynb: check_stationarity function and ADF test output

Data-Mining-Accident-Severity/Data Mining_Accident_Severity.ipynb: confusion matrix and explicit sensitivity calculations for Severity 4

Data Wrangling & Cleaning

6/10

Preparing and cleaning data

Solid data wrangling at notebook scale: large CSVs are parsed, datetimes converted, null columns identified and dropped, feature extraction (street type) and dummy encoding are implemented; imputation choices are pragmatic but somewhat ad hoc (ffill used broadly) and there is limited provenance or defensive checks against train/test leakage.

Evidence

Data-Mining-Accident-Severity/Data Mining_Accident_Severity.ipynb: drop columns with >40% null and df = df.fillna(method="ffill")

Data-Mining-Accident-Severity/Data Mining_Accident_Severity.ipynb: Street type extraction loop creating Street2

Time-Series-Predictive-Analytics/Time-SeriesPredictive Analytics.ipynb: conversion of Date to datetime and grouping to daily/weekly counts

Exploratory Analysis & Visualization

6/10

Exploring and visualizing data

Exploratory analysis is thorough with many targeted plots and accompanying textual interpretations; visuals are used to drive decisions (sub-dataset selection, seasonal checks), though some plots are standard templates and deeper causal interrogation is limited.

Evidence

Data-Mining-Accident-Severity/Data Mining_Accident_Severity.ipynb: multiple sns.countplot and boxplot sections with short written interpretations (Severity, month, hour, State plots)

Time-Series-Predictive-Analytics/Time-SeriesPredictive Analytics.ipynb: ACF/PACF plotting and stationarity visualizations with written commentary

Predictive Modeling

5/10

Building models that predict

Predictive modeling is applied end-to-end with reasonably large-scale experiments (DecisionTree, RandomForest, SARIMAX, auto_arima, rolling forecast); however models lack systematic hyperparameter tuning, rigorous cross-validation (classification uses a single train/test split), calibration, and deeper error analysis beyond confusion matrices and RMSE summaries.

Evidence

Data-Mining-Accident-Severity/Data Mining_Accident_Severity.ipynb: DecisionTreeClassifier and RandomForestClassifier training and test evaluation with confusion matrices

Time-Series-Predictive-Analytics/Time-SeriesPredictive Analytics.ipynb: SARIMAX modeling, auto_arima usage and rolling forecast implementation

Business Insight & Impact

4/10

Turning analysis into business value

Some attention to business impact is visible (explicit concern about low sensitivity for severity-4 accidents and converting to a binary target to raise detection), but there is limited quantification of costs of false positives/false negatives or operational trade-offs and no clear deployment/monitoring plan tied to business metrics.

Evidence

Data-Mining-Accident-Severity/Data Mining_Accident_Severity.ipynb: discussion and calculations on Severity 4 sensitivity and the decision to convert to a binary target to improve detection

Time-Series-Predictive-Analytics/Time-SeriesPredictive Analytics.ipynb: 'Risks or Ethical Concerns' section and RMSE/P-RMSE evaluation framing

Reproducibility & Notebook Hygiene

3/10

Clean, repeatable analysis

Notebooks are executed and contain narrative, but reproducibility and engineering hygiene are limited: no pinned environment, no requirements.txt or environment spec, seeds are inconsistently set, and work is notebook-first rather than packaged for production.

Evidence

Data-Mining-Accident-Severity/Data Mining_Accident_Severity.ipynb: use of DecisionTreeClassifier(random_state=0) but RandomForestClassifier() instantiated without random_state and no requirements/environment file

Time-Series-Predictive-Analytics/Time-SeriesPredictive Analytics.ipynb: executed modeling workflow with no environment or reproducibility artifacts (no pinned deps or DVC/data versioning)

Industries

Government· Middle

Transportation & Logistics· Middle

Technologies

Jupyter Notebook

Scikit-learn

Seaborn

Matplotlib

Pandas

NumPy

Statsmodels

Time Series

Analytics· mentioned only

Recommendations

Productionize forecasting and classification workflows: package notebooks into repeatable scripts or a lightweight pipeline, pin dependencies (requirements.txt or environment.yml), and add data/versioning (DVC or similar).
Improve model validation: add cross-validation where appropriate, systematic hyperparameter tuning (Optuna/GridSearch), and calibration or threshold optimization for imbalanced class detection (severity 4).
Harden data engineering: replace broad forward-fill imputation with feature-specific strategies, add provenance and unit tests for data transforms, and guard against train/test leakage.
Add model interpretability and monitoring: compute feature importances, SHAP analyses for critical predictions, and define business metrics that quantify costs of false negatives vs false positives for severity alerts.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

arash-mahmoudian Data-Mining-Accident-Severity Time-Series-Predictive-Analytics

Job preferences

Work permits: United States

Open to AI agents: they read this profile and start a dialog through Alionalion.io/@arash-mahmoudian/mcp