This week marked significant strides in our Verifone project. Each team delivered key accomplishments, moving us closer to our goals. The collaboration between the UI, Feature Engineering (FE), and Machine Learning (ML) teams demonstrated remarkable synergy as we tackled core components and refined functionalities across the board.

The UI Team reached a significant milestone by completing the site selection of core visual components and their integration. While site-specific statistics remain in progress, clustering sites on the map has been implemented, providing a more intuitive way to visualize the data. Additionally, the search bar is fully functional, seamlessly integrated with the map, and includes improved statistics details to enhance usability.

The Feature Engineering Team shifted its focus to feature correlation analysis, which will help us identify and limit feature redundancy. The team also made significant headway in building the pipeline to simulate multi-site data live streams, ensuring a reliable feed into the broader data processing pipeline.

On the Machine Learning Team, spatial models saw a breakthrough with the successful implementation of Gaussian Mixture Models (GMM) and DBSCAN. The team also fixed the long-term memory issues in the CNN-LSTM model and completed the ensemble pipeline. Synthetic data testing and debugging were conducted to validate these models, paving the way for more accurate predictions and insights.

As we prepare for the coming week, each team has outlined clear goals to maintain momentum. The UI team will finalize site-specific statistics and explore implementing filters and dynamic cluster/marker colors based on volume. The Feature Engineering team will wrap up feature correlation analysis and work on reformatting to reduce preprocess runtimes. Meanwhile, the ML team plans to isolate and benchmark different ensemble models, fine-tune atomic models, and adjust ensemble weights.

We are set for another productive week with consistent progress across all teams. Let’s keep up the excellent work! All teams, we are set for another productive week. Let’s keep up the excellent work!

This week marked significant progress for our Verifone project, with each subteam making meaningful strides toward our goals. The UI team completed the core visual components of the main screen, providing a solid foundation for displaying key data. They also began backend integration, including mapping site displays, statistics visualization, and connecting model outputs with the UI. These efforts bring us closer to a cohesive and functional user interface.

The Feature Engineering (FE) team focused on preparing JSON data files for batch processing, a crucial step for handling larger datasets efficiently. They explored new ideas for feature engineering other event types, broadening the scope of our model’s potential applications. Additionally, they worked on optimizing the preprocessing pipeline for model training and prediction, ensuring smoother integration across the project.

The Machine Learning (ML) team made significant advancements this week, starting with extracting anomalies from new data to present actionable results for the UI. They also generated synthetic data to test and validate their models, calculating baseline metrics such as F1 scores for the ensemble design. Furthermore, prototypes of classification models, including linear and logistic regression, were completed, establishing a framework for the next stage of development.

Next week, each subteam has outlined clear objectives to build on this week’s momentum. The UI team will complete the visual components for the site popup screen, refine statistics displays for the primary and popup screens, and enhance functionality by improving the search bar and filters. The FE team will focus on implementing a caching system to track PREAUTH sequences and redesigning the build sequence function to handle multiple sites. The ML team plans to explore unsupervised learning models such as DBSCAN and GMM, finalize the ensemble design, and fine-tune each atomic model and ensemble weights to optimize performance metrics.

With steady progress across subteams and clearly defined plans, we are well to achieving our following milestones. See you next week with more updates and accomplishments!

This week, our team made significant progress across all fronts, refining our pipeline and strengthening our model’s capabilities. As a team, we successfully presented QRB 1 and began preparing slides for QRB 2 and the FDR documentation. Meanwhile, work continued programming the UI frontend, ensuring a seamless and user-friendly experience.

The UI team took a significant step forward by initiating the programming pipeline, successfully automating and integrating it with the model output. This achievement streamlines the process, making it more efficient and scalable. On the feature engineering (FE) side, the team reworked the CHAMPS preprocessing scripts, optimizing them for smoother data handling. Additionally, they refined the training process by allowing more flexibility in using feature engineering columns, ensuring more adaptable model training.

The machine learning (ML) team also made substantial advancements, introducing an ensemble anomaly matrix and developing an ensemble batch model training script. These enhancements improve anomaly detection by leveraging multiple models for more robust predictions. Another key milestone was the automation of synthetic data injection, making it easier to test and validate models under controlled conditions. The team also formatted the output for anomaly detection, ensuring consistency and clarity in results.

Further strengthening our dataset, the FE team incorporated additional spatial features, including region segmentation, population size, gender ratio, and median household income. These enhancements provide more contextual data, helping the model make more informed predictions.

Our project is becoming increasingly refined and automated with these developments, bringing us closer to a fully functional and efficient anomaly detection system. We’re looking forward to continuing this momentum in the coming weeks. See you next time!

This week, Team INSIGHT continued making strides in our Verifone project, collaborating to complete the required QRB documents. These efforts ensured that we were aligned with Verifone’s expectations and prepared for the subsequent phases of development. Each subteam made notable progress, laying a strong foundation for upcoming tasks.

The UI team focused on designing the whole pipeline’s backend architecture, incorporating Verifone’s suggestions. This design work is critical to ensuring a seamless integration of the front and back ends. The Feature Engineering (FE) team implemented z-score calculations for duration and amount using a rolling window, experimenting with different designs to optimize runtime. They also made significant progress in spatial feature engineering, extracting key geographical attributes such as city, state, and zip code. The Machine Learning (ML) team advanced their work on isolation forests, seq2seq models, and Optuna integration, further refining our approach to anomaly detection.

Looking ahead, the subteams have ambitious plans for next week. The UI team will continue Angular development, focusing on map integration and clustering research while finalizing anomaly and statistics scripts for UI implementation. The FE team aims to complete their proposed spatial features and optimize the preprocessing pipeline in collaboration with the ML team. The ML team will work on formatting JSON outputs for backend integration, ensuring that anomaly detection results are effectively communicated.

As we reflect on this week, it’s clear that collaboration and communication remain the keys to our success. We are excited to build on this momentum as we move closer to delivering a polished and impactful solution.

As we embark on a new semester, the excitement within Team INSIGHT is palpable. After a refreshing winter break, we are all rejuvenated and ready to dive back into the Verifone project with fresh energy and ideas. This week, we reconnected with one another and met with our coach to align our goals and plans for the semester. It was an excellent opportunity to reflect on past progress and set the stage for the exciting work ahead.

Our team hit the ground running this week, achieving significant milestones across all subteams. The UI team began developing the user interface, familiarizing themselves with Angular’s development cycle, and setting up the backend for the site. This included configuring the Mongo Atlas database and testing HTTP methods, marking a strong start toward building a robust and dynamic platform.

The Feature Engineering (FE) team made notable strides in data preprocessing by implementing cyclical and binary encoding methods. These refinements enhanced the timeline for data preparation, ensuring smoother downstream processes.

The Machine Learning (ML) team also had a productive start, refining the prediction script and enhancing the model auto-search functionality. These improvements lay the groundwork for more accurate anomaly detection and efficient model integration moving forward.

Looking ahead, each subteam has a focused plan to build on this momentum. The UI team will continue their Angular development efforts, emphasizing integrating a map and researching clustering methods. They will also finalize anomaly and statistics scripts for UI implementation. The FE team plans to complete the proposed features, including normal, z-score, and spatial grouping. In contrast, the ML team will work on formatting JSON outputs and integrating Isolation Forest and Optuna for improved anomaly detection.

We are thrilled about the progress made in this first week and eager to continue building on it in the coming weeks. Here’s to a semester filled with innovation, collaboration, and success!

Goodbye for now, and see you next week!