Boston property assessment dataset from Boston.gov classifies properties in greaterboston area into it’s present overall condition (Poor to Excellent).
4 classification algorithms (Naives Bayes, RandomForest, IBk and Decision Table) were modeled using 5 different selection attributes using Weka.
Performance measures such as TP Rate, FP Rates, ROC Area etc were used to determine the overall performance of each classifier model.
Project page : View here
1. The project analyses the MBTA real-time data of 1 week data between Dec 1- Dec 8.
2. Advanced Data visualization of Travel times and Head way times are implemented with R and Plotly in-order to determine the problems that exist in the Boston subway lines. The Intend of doing this was to develop a weekly reporting app that can be used to determine weekly estimates of the issues that needs to be looked upon by the MBTA.
3. A real-time MBTA app is developed using R and shiny. This app also shows real-time clustering of the trains based on parameters such as subway line length, number of active trains in a direction, and the distance to the next nearest train etc.
4. The shiny app provides a simple general population facing app where anyone can keep track of the train locations with the simplest of UI.
5. The shiny app from the MBTA facing side, predicting/visualizing train clusters and giving a birds eye dashboard in each train so that the driver would be able to adjust the train speed in-order to avoid clustering. This should at-least reduce the frequency of trains going in express mode and make the waiting time for trains at stations much more uniform. Usually, the cluster keeps building up until someone from the control room calls up the driver of the specific train (as the driver is unaware that a train is following till he is contacted from the control room) to move express to certain stop.
Project page : View here
We use Expected Transmission Time (ETT) as the metric for controller hand-off inOpenFlow WMNs. ETT reflects various physical-layer characteristics, such as link traffic and end-to-end bandwidth. The experimental results showed that ETT is a better metric compared to RTT and ETX in a dynamic network with variable load across the links. ETT-based hand-off is able to respond to the excessive load in the link and make suitable hand-off decisons, whereas RTT and ETX fails in accomplishing the same with lower hand-off delay and packet dropouts.
Paper : View here
Software Defined MICRONet architecture provides intelligent communication among physical boat clusters in the sea. This will solve the technology challenges faced by the fishermen community in India today, specifically by providing software defined Intelligent and adaptable communication and connectivity while they are out at sea. A scaled down model of Software Defined MICRONet environment was emulated in a testbed.
Final Report : View here
Repository : Software Defined MICRONet (2016)
Head Mounted Display(HMD) is one of the revolutionary Virtual Reality(VR) inventions of all times. But how do you move around in a Virtual Environment?. For a true VR experience you need to move around freely and naturally. Imagine a game where the user can freely roam around their backyard or walk on a frictionless surface and navigate in a virtual environment rather than sitting idle in a chair. Developing a low-cost system for such a VR experience which can be implemented onto a HMD, is always a challenge. In this project we have done a hardware implementation to navigate in a virtual environment using a low-cost Inertial Measurement Unit(IMU).
Repository : Navigation in a Virtual Environment using IMU MPU-6050 (2015)
Developed a 32 bit RISC Microprocessor in VHDL language and implemented on Altera FPGA. The Test bench module is executed in the model-sim software and the LCD module is implemented on the FPGA to display the Register value, Memory value and the Program counter.
Repository : 32 bit RISC Microprocessor in VHDL language and implemented on Altera FPGA (2014)
For each input pixel window, the values in that window are multiplied by the convolution mask. Next, those results are added together and divided by the number of pixels in the window. This value is the output for the origin pixel of the output image for that position. The input pixel window is always the same size as the convolution mask. The output pixel is rounded to the nearest integer. The results for this algorithm carried over an entire input image will result in an output image with reduced salt-and-pepper noise. This flexibility allows for many powerful uses.