A

Describe Your Problem

Enter your machine learning problem or dataset goals below and click 'Analyze & Recommend' to generate server-side Gemini AI model suggestions.

How this works
  1. Type or paste your project goal or problem description in plain text (up to 750 words).
  2. Click "Analyze & Recommend" to send your problem statement to backend Gemini AI.
  3. You will automatically navigate to Model Guidance to review recommended model algorithms.
Problem Statement
0 / 750 words
Describe → AI Analysis

Model Guidance

Review AI model evaluations and click 'Make this simpler' to ELI5 technical jargon into plain language, or hover terms for definitions.

How this works
  1. Review the top recommended machine learning algorithms selected specifically for your problem statement.
  2. Click "Make this simpler" repeatedly — each click makes the explanation noticeably simpler down to an ELI5 child analogy.
  3. Hover over highlighted technical terms (dotted purple underline) for instant glossary definitions.
AI Model Recommendation Summary
Submit a problem description on the Describe page to generate custom AI recommendations.
Recommended Model Family

Dataset Cleaning Engine

Upload up to 3 raw tabular datasets simultaneously, select outlier cleaning methods (IQR vs Z-Score), and download processed files in their original formats.

How this works
  1. Select up to 3 dataset files (.csv, .json, .tsv, .parquet, .txt) in the 40% opaque light yellow upload cards below.
  2. Choose outlier detection method on the right bar: IQR (Interquartile Range) or Z-Score (standard deviations). Hover for 20-word explanations.
  3. Click "Clean Dataset" or "Clean All 3 Datasets" to generate step-by-step audit reports and download cleaned files in original format.
1 Dataset Upload Box #1
Outlier Method:
📊 IQR (Interquartile Range)
IQR measures spread of the middle 50% data between 25th and 75th percentiles ($IQR = Q3 - Q1$). Outliers are marked outside $[Q1 - 1.5 \cdot IQR, Q3 + 1.5 \cdot IQR]$.
📈 Z-Score Standard Deviation
Z-Score standardizes values measuring how many standard deviations ($z = (x - \mu)/\sigma$) a point lies from the mean, flagging outliers where $|z| > 3.0$.
2 Dataset Upload Box #2
Outlier Method:
📊 IQR (Interquartile Range)
IQR measures spread of the middle 50% data between 25th and 75th percentiles ($IQR = Q3 - Q1$). Outliers are marked outside $[Q1 - 1.5 \cdot IQR, Q3 + 1.5 \cdot IQR]$.
📈 Z-Score Standard Deviation
Z-Score standardizes values measuring how many standard deviations ($z = (x - \mu)/\sigma$) a point lies from the mean, flagging outliers where $|z| > 3.0$.
3 Dataset Upload Box #3
Outlier Method:
📊 IQR (Interquartile Range)
IQR measures spread of the middle 50% data between 25th and 75th percentiles ($IQR = Q3 - Q1$). Outliers are marked outside $[Q1 - 1.5 \cdot IQR, Q3 + 1.5 \cdot IQR]$.
📈 Z-Score Standard Deviation
Z-Score standardizes values measuring how many standard deviations ($z = (x - \mu)/\sigma$) a point lies from the mean, flagging outliers where $|z| > 3.0$.

Dataset Supply

Browse curated machine learning datasets below and click 'Download' to stream raw files directly to your device.

How this works
  1. Search or filter curated machine learning datasets from the supply library.
  2. Click "Download" to stream raw dataset files straight to your computer.
  3. New datasets are curated and added to the supply library directly by developers.

Testing Framework

Select a model estimator (or upload your custom model file) and click 'Execute Testing Pipeline' to evaluate performance and view accuracy/R² metrics.

How this works
  1. Optionally upload your own custom model artifact (.py, .js, .pkl, .html) using the box below.
  2. Select an estimator algorithm and specify the target task type (Classification vs. Regression).
  3. Click "Execute Testing Pipeline" to run model evaluation and inspect metrics and logs.

You can input models of types - .py/.js/.pkl/.html

Evaluation Config
Random Forest
🌲 Random Forest
Random Forest creates a large forest of independent decision trees during training. When making predictions, each tree casts a vote, and the majority decision wins. This ensemble approach prevents individual trees from memorizing noise (overfitting) and delivers high stability across varied real-world data types.
XGBoost
⚡ XGBoost
XGBoost (eXtreme Gradient Boosting) is an advanced machine learning algorithm that builds decision trees sequentially. Each new tree focuses on correcting errors made by previous trees. It is world-renowned in Kaggle competitions for delivering top prediction accuracy on tabular CSV datasets while executing with exceptional computational speed.
Neural Network
🧠 Neural Network
A Neural Network (Multi-Layer Perceptron) mimics the human brain using layers of interconnected nodes called neurons. Inputs pass through hidden layers with weighted connections to learn complex, non-linear relationships. It excels at recognizing deep patterns in intricate datasets, images, and unstructured features.
Decision Tree
🌿 Decision Tree
A Decision Tree breaks down data by asking a sequence of simple yes-or-no questions, creating a tree-like flowchart of decision rules. It is exceptionally easy to visualize and interpret, making it an ideal beginner-friendly baseline model for understanding feature splits.
Logistic Regression
📈 Logistic Regression
Logistic Regression is a foundational statistical model that calculates probabilities for classification outcomes. It measures the relationship between input features and target categories using a smooth S-shaped sigmoid curve, providing a fast, highly interpretable baseline for binary or multi-class predictions.
SVM
📐 Support Vector Machine (SVM)
Support Vector Machine finds the optimal decision boundary (hyperplane) that maximizes the margin distance between different data classes. Using mathematical kernel functions, SVM can project complex non-linear data into higher dimensions to cleanly separate difficult overlapping classes.

Training Framework

Upload a trained model file, configure cross-validation folds, then click 'Start Model Training' to execute model training and inspect performance logs.

How this works
  1. Optionally upload your own custom model artifact (.py, .js, .pkl, .html).
  2. Choose cross-validation split strategy (5-fold vs 10-fold CV).
  3. Click "Start Model Training" to execute fitting loops and monitor execution progress.

You can input models of types - .py/.js/.pkl/.html

Training Options
5-Fold Cross Validation
📊 5-Fold Cross Validation
5-Fold Cross Validation splits your dataset into 5 equal parts. The model trains on 4 parts and tests on the remaining 1 part, repeating this process 5 times so every data point is tested once. This gives a reliable estimate of model performance without risking lucky split results.
10-Fold Cross Validation
🔄 10-Fold Cross Validation
10-Fold Cross Validation divides your dataset into 10 equal sections. The algorithm trains on 9 sections and tests on the remaining 1 section, repeating this 10 times. It provides an even more robust and thorough accuracy benchmark than 5-fold CV, especially for larger datasets.
XGBoost
⚡ XGBoost
XGBoost (eXtreme Gradient Boosting) is an advanced machine learning algorithm that builds decision trees sequentially. Each new tree focuses on correcting errors made by previous trees. It is world-renowned in Kaggle competitions for delivering top prediction accuracy on tabular CSV datasets while executing with exceptional computational speed.
Random Forest
🌲 Random Forest
Random Forest creates a large forest of independent decision trees during training. When making predictions, each tree casts a vote, and the majority decision wins. This ensemble approach prevents individual trees from memorizing noise (overfitting) and delivers high stability across varied real-world data types.
Neural Network
🧠 Neural Network
A Neural Network (Multi-Layer Perceptron) mimics the human brain using layers of interconnected nodes called neurons. Inputs pass through hidden layers with weighted connections to learn complex, non-linear relationships. It excels at recognizing deep patterns in intricate datasets, images, and unstructured features.

User Profile & Activity

Manage your account profile details and view historical machine learning problem statements entered in the Describe box.

A

User Name

user@datadealers.local
DataDealers Member
Describe Box Problem History 0 Problems Entered

About DataDealers

Unified AI/ML Data Engineering & Model Guidance Platform — Interdisciplinary Project Report (CSE-8)

Platform Overview

Data Dealers is a unified, beginner-accessible web platform designed to guide engineering students, hackathon participants, and early-career developers through every phase of the AI/ML project lifecycle. It combines structured learning material, automated dataset cleaning, intelligent model recommendation, and testing into one seamless guided workflow.

Core System Modules

1. Learning Materials Module

Provides easy-to-understand guides covering fundamental AI data concepts, labeling conventions, and common pitfalls, creating an educational foundation for beginner practitioners.

2. Dataset Discovery Module

Enables users to quickly discover, filter, and stream curated, AI-ready datasets tailored to their specific project goals, eliminating tedious search across general-purpose repositories.

3. Model Guidance Module

Accepts plain-language project objectives and data characteristics to recommend optimal machine learning algorithms, input data formats, and training strategies.

4. Data Cleaning Module

Automatically cleans raw uploaded datasets by handling missing values and outlier detection via IQR or Z-Score statistical methods, generating structured, ready-to-train files.

5. Testing & Training Framework Module

Offers interactive model testing, custom model artifact evaluation, and k-fold cross-validation execution while surfacing execution logs and precision metrics.

----- Teammates ----- Students of CSE-8 (SEM 2) 1)Aditya Rohan Chourasia - (1TD25CS016) 2)Sooraj G. - (1TE25CS284) 3)Arya Shetty - (1TE25CS040) 4)Bhuvanesh D. - (1TE25CS069)