Mourad Askar's repositories
chicago_traffic_crashes
This study creates machine learning models to predict the seriousness of car crashes using 2019 and 2020 crash reports from the publicly accessable database maintained by the Chicago Police Department. A car crash is considered serious if the crash results in an injury or the car is towed due to the crash. Models use categorical features that describe conditions at the time of the crash and crash causes to predict the required target. The current focus is to classify whether a crash results in an injury. All machine learning models are trained, validated, and tested on randomly split 2019 crash reports. The best model (along with all others) are then tested using the full set of 2020 crash reports.
conda_configs
My conda environments configurations
credit_card_churners
Customer relations are essential for businesses to keep up proactive service brand and maintain a high standard of communication. In banks, credit card customers churn, and banks lose business. The dataset of bank credit card customers' churn provides many customer profile characteristics. They list cardholders' profiles and capture their tiers of credit limits, account age, and other activity and demographic features. We analyzed the factors that affected the customers' attrition and developed models to predict and identify customers at high risk for attrition so that the bank management can understand the potential reasons for churning customers and proactively act upon them. The study intends to provide support to the customer success division in banks to increase customer engagement and allow customer relations to interact with those high-risk customers to maintain them proactively. The dataset is from the Kaggle website. It has 10K records and 20 features. 15 numerical and 6 categorical. The dataset was complete and did not have any missing values, and it did not require cleaning. The target feature of interest is a binary flag indicating whether a customer is an attrited customer or not. The target feature is imbalanced at 16% ratio. To counter the imbalance, we applied resampling and class weight factors attributes. The dataset is split into training and testing subsets. The training, tuning, and validation steps used the training dataset, and the testing dataset was held out to be only used in the last step to test the models. Multiple machine learning models were evaluated, including Stochastic Gradient Descent, Decision trees, and Random Forests. The best results were achieved by the Random Forest ensemble achieving 0.95 accuracy while maintaining high scores for F1, Recall, and Precision (0.86, 0.90, 0.83). The study concluded that account activity-related features such as the amount and count of transactions were the top influencing features that affected the attrited customers and that demographic features were not helpful.
init_linux
Scripts to initialize new linux environments with my preferred settings
metro_interstate_traffic_volume
The project aims to predict the volume of traffic flow between Minneapolis and St. Paul at a specific point in Minnesota. My task is to build a multi-step RNN with LSTM model that makes a single prediction point of the traffic volume 2 hours into the future, given the previous 6-hour window.
scripts_shell
My shells scripts
socrata_public_data_api
A walkthrough to access public data using socrata API
us_pres_polls_2020
US Presidential Polls 2020 (Trump vs. Biden)
stock_market_time_series
Applying time series analysis to a few stock tickers