Projects
Employee Attrition Problem Analysis and Prediction
This project solves the problem of employee attrition which reaches more than 10% using in-depth analysis with output in the form of a business dashboard and attrition predictions using machine learning, as well as several recommended action items that can be carried out by companies.
Student Dropout Problem Analysis and Prediction
This project solves the problem of student dropout which is quite high, reaching more than 32% using in-depth analysis with output in the form of a business dashboard and student status predictions using machine learning, as well as several recommended action items that can be carried out by the institution.
Stroke Disease Detection
Developed and deployed a stroke detection model using Predictive Analytics and MLOps, achieving 96% accuracy. Evaluated performance with AUC, Binary Accuracy, TFMA metrics, and confusion matrix components. Focused on improving early detection for high-risk individuals, enhancing healthcare insights through machine learning-driven diagnostics.
Disaster Tweets Classification
Developed and deployed a disaster tweets classification model using Natural Language Processing (NLP) and Machine Learning Operations (MLOps), achieving 86% accuracy. Evaluated performance with AUC, Binary Accuracy, TFMA metrics, and confusion matrix components. Focused on detecting fake news or hoaxes that can quickly spread on widely used platform such as Twitter about natural disasters.
Bike Sharing Analysis Dashboard
This project is part of the bike-sharing data analysis project to analyze the Bike Sharing Dataset. The results of the analysis are then made into the form of data visualization into an interactive dashboard.
Book Recommender System
Developed an advanced book recommendation system to reignite reading interest, leveraging data from AWS. Implemented Content-Based Filtering with TF-IDF and Cosine Similarity, alongside a custom Collaborative Filtering model using RecommenderNet with Binary Cross-entropy and Root Mean Squared Error (RMSE) metrics. Optimized data preprocessing and analysis to enhance recommendations. Notable insights include peak book demand in December (12%) and the lowest in June (6%).
Electric Predictive Analytics
Developed a machine learning model to predict electricity consumption in Tétouan, Morocco, using weather data. Analyzed 52,416 observations from three zones, identifying key correlations. Random Forest outperformed other models, achieving an Root Mean Squared Error (RMSE) of 24.15 (train) and 39.28 (test). This project addresses energy efficiency challenges and supports sustainable resource management.
Chicago Weather Forecasting
Developed an LSTM-based deep learning model for weather time series prediction in Chicago using 43,824 data points. Trained with Stochastic Gradient Descent (SGD) optimizer, Huber loss, and Mean Absolute Error (MAE) metric, achieving a MAE of 2.2306 and validation MAE of 1.7385. Also implemented early stopping for model training optimization.
News Classification with NLP
Developed a deep learning model using Bi-LSTM for news topic classification (world, sports, business, sci-tech) on 120,000 data points. Applied data cleaning, trained with Adam optimizer and categorical cross-entropy loss, achieving 97.27% accuracy. Implemented ReduceLROnPlateau for model training optimization.
Rock-Paper-Scissors Image Classification
Developed a deep learning model using a Convolutional Neural Network (CNN) to classify rock, paper, and scissors hand images. Trained on 2,188 images (60:40 split) with the Adam optimizer and categorical cross-entropy loss. Achieved 97.62% accuracy and 98.75% validation accuracy in 20 epochs.