Hello, I'm Syed Tanveer.Machine Learning Developer Experience in Apache Spark, Python and TensorFlow. Ocassionaly works with Scala and R.
About Me
I am passionate about Artificial Intelligence. Currently, I am building machine learning applications at an Ad Tech company. I have received a Masters in Computing Science from Simon Fraser University in Canada and a Bachelors in Computer Science with the highest distinction Summa Cum Laude from North South University in Bangladesh. On a random note, I am a huge fan of the Chelsea Football Club.
I am interested in building systems which can process large-scale data and can be employed to take intelligent decisions. I am intrigued by the idea that how we use computers to surpass the human limits in designing the solutions of the problems that exceed our understanding. (Google: Predpol)
Some of the areas I am interested in are listed below:
Recommendation System
Text Mining
Deep Learning
Business Intelligence
Publication List
Published
Journal
Jishan,
S. T., Rashu, R. I., Haque, N., & Rahman, R. M. (2015).
Improving accuracy of
Students' Final Grade Prediction Model using Optimal Equal Width
Binning and Synthetic Minority Over-Sampling Technique. Decision
Analytics, 1-24.
Published
Conference Papers
Kamal,
A., Jishan, S.T. ,Monsur,N. & Ahmed,N. ChaScript:
Breaking Language Barrier using a Bengali Programming System.
In 8th International Conference on Electrical and Computer
Engineering, Dhaka, Bangladesh, December 20-22, 2014.
Published
Book Chapter
Jishan,
S. T., Rashu, R. I., Mahmood, A., Billah, F., & Rahman, R. M.
(2015). Application
of Optimum Binning Technique in Data Mining Approaches to Predict
Students’ Final Grade in a Course. In Computational Intelligence
in Information Systems (pp.
159-170). Springer International Publishing.
Estimating Force and Torque exerted by hand based on FMG Signals using Generalized Regression Neural Network
A machine learning model was necessary to estimate force generated by hand for robot to replicate the human action as there is no first principle law in physics to translate between force myography signals and force generation. Generalized Regression Neural Network was used along with Autoencoder for dimensionality reduction. A favorable R2 score (0.82) was achieved after several trial and error.
Personalized Temporal Recommender System using Recurrent Neural Network
Recommender systems are used to suggest products to audiences by employing a similarity metric. One of the problem of such systems is that it does not incorporate the context of time. As result, it is not possible to change recommendation as audiences' preferences changes over time. In this paper, we will be presenting a solution based on recurrent neural network to alleviate this problem and highlight a use case on how recurrent neural network model can help us build a real-time recommender system.
Crime data analysis is fundamental to understanding crime patterns which will aid in preventing future crimes to happen. In this project I have build a data warehouse of Vancouver crime data using Microsoft SSAS and Tableau which will allow users to make precise queries. Furthermore, I have provided an interactive geo-visualization system for better understanding of crime situation. I have also commenced analysis on this data to answer some of the important questions related to crime in Vancouver.
Improving Item coverage for recommendation in disjoint social network
Collaborative filtering (CF) based recommender systems suggest items to users by employing a similarity
metric. With the introduction of online social networks, graph-based approach of recommendation has emerged. This approach assumes trust among users and recommends items based on trust. Although it solves cold start problem in recommendation system, user-item coverage decreases. We argue that one of the causes of coverage issue is the disjointedness of trust network, therefore, an item not rated by anyone in a subgraph cannot be recommended to any user in that subgraph. To solve this problem, we propose TrusTem: a recommendation algorithm that merges user-item and trust domains. We have conducted experiments on real-life dataset and compared TrusTem againstmstate-of-art algorithms. Our experiments demonstrate that TrusTem covers 99.8% of user-item pairs, at cost of reducing accuracy up to 8%. However, we show numerically that TrusTem does better coverage-accuracy trade-off than these algorithms.
- System built using Node.js, designed to work like a human news editor for the citizen journalism platform.
- It can extract current trends in Bangladesh and filter out the breaking news through time series analysis.
- Fuzzy Inference System to understand article quality.
- Unsupervised Learning for the selection and allocation of the articles.
As technology is evolving as a driving force in our
society, the need for computer programming is arising.
However, since most of the computer programming languages
are English-based, they can act as barriers for learning
computer programming for people who are not adequate in
English. We have build a Bengali programming system with the goal to help Bengali speaking people learn and write computer programs.
Optimum Binning + SMOTE based Grade Prediction Models
There is a perpetual elevation in demand for higher education in the last decade all over the world; therefore, the need for improving the education system is imminent. Educational Data Mining is a newly-visible area in the field of Data Mining and it can be applied to better understand the educational systems in Bangladesh. In this research, we present how data can be preprocessed using a discretization method called the Optimal Equal Width Binning and an over-sampling technique known as the Synthetic Minority Over-Sampling (SMOTE) to improve the accuracy of the students' final grade prediction model for a particular course. In order to validate our method we have used data from a course offered at North South University, Bangladesh. The result obtained from the experiment gives a clear indication that the accuracy of the prediction model improves significantly when the discretization and over-sampling methods are applied.
Kajer Khoj
Job Search Engine for Bangladesh. Built using PHP framework CodeIgniter with Bootstrap on the frontend for responsive design.
The idea behind the development of SOS BD is to provide people of Bangladesh with a sophisticated mobile application that will come in handy during emergency situations. This application minimize the communication gap between the people and the emergency services such as Ambulance, Fire Service etc. Finding the global position of the user and using that to calculate the nearest emergency services available for the user is the main goal of the application.Furthermore, “Quick SMS” service is being integrated so that the user can send text messages to certain phone numbers, automatically stating his/her location along with global position.