m
muhamdimrn

Muhamad Imran

@muhamdimrn

Grow your business with me!

Malaysia
English, Malay
About me
Hi! I'm Muhamad Imran, a Python developer and data analyst from Malaysia specializing in machine learning, data pipelines, Power BI dashboards, and data entry. I build predictive models using XGBoost and deep learning, automated data pipelines, and interactive dashboards that turn numbers into clear decisions. Certified by Google, IBM, and Microsoft (Power BI). I have hands-on experience building real ML systems with automated prediction pipelines and Telegram integrated bots. Whether you need data entry, Python scripts, or a Power BI dashboard, I am ready to help.... Read more

Skills

m
muhamdimrn
Muhamad Imran
Offline • 
Average response time: 1 hour

See my services

Data Dashboards
I will build you a professional power bi dashboard
From$40 / projectMore details
Time Series Analysis
I will build an xgboost time series prediction model
From$90 / projectMore details

Portfolio

Work experience

Freelancing_Career

Machine Learning Engineer

Freelancing Career • Self-employed

Dec 2025 - Present • 6 mos

Developed a full end-to-end algorithmic trading system for XAUUSD (gold) using XGBoost to predict session High and Low prices across three global trading sessi…Developed a full end-to-end algorithmic trading system for XAUUSD (gold) using XGBoost to predict session High and Low prices across three global trading sessions: Asian, London, and New York. The system was built on 7 years of M15 candlestick data (2019 to 2026) using a 7-round walk-forward validation framework to prevent data leakage and ensure the model generalizes to unseen market conditions. Each round trained on historical data and tested on the following year, simulating real-world deployment conditions. Feature engineering involved 28+ technical indicators including ATR-normalized targets, momentum indicators, volatility measures, pivot point distances, Bollinger Band Width Percentile, and session-based range features. Features were carefully selected per session to maximize predictive power while avoiding redundancy. Hyperparameter tuning was performed using Optuna with 200 trials per model, optimizing for minimum MAE on the validation set. A meta-learner (Model C) was built using XGBoost to correct predictions from the base model, achieving a 15.44% MAE improvement over the baseline. Final model achieved MAE below 0.50% across all sessions. The complete pipeline was integrated with a Telegram bot for automated signal delivery, sending trade signals including entry price, take profit, and stop loss levels at the start of each trading session. The project involved building a multi-model ensemble architecture including a direction classifier (XGBoost), a deep learning TCN model for High/Low prediction, and a meta-learner for prediction correction. Full pipeline covered data collection, cleaning, feature engineering, model training, evaluation, simulation, and live deployment. Tools: Python, XGBoost, TensorFlow, Optuna, Pandas, NumPy, Matplotlib, Scikit-learn, Telegram Bot API