this is the types of problems in Machine Learning

problems_in_ml.md

types of problems in machine learning

• general

  • how to collect data
  • What type of ML problem is this
  • what is the type of data
  • what will be the output
  • missing values and outliears
  • labeled or unlabeled data
  • what kind of preprocess and transformation needed
  • potential bieases in the model
  • what is the size of training and tesing data
  • possibiliy of data leakage
  • overfit and underfit chances
  • How to evaluate the model
  • what is the simplest way to make a model
  • how is to be diployed in the real world
  • how to monitor and update model in production
  • What ensemble methods can i consider
  • How to explain the model predictions
  • model serving and scalability
  • how to update the model with new data

• regression

  • if linear relation
  • if polynomial relation
  • if some multicoliniarity
  • if future selection required
  • for count data
  • if a lot of non linear relations
  • outliers
  • missing values
  • if feauture importance is crutial
  • if homoscedasticity is violated
  • if the feautures are high dimention
  • if real time prediction required
  • if regularization needed
  • If the data distribution is skewed
  • Categorical data included
  • interpretebility is important
  • if training time is constraint
  • Data has temporal dependencies
  • Interacion effects

• classification

  • Missing values
  • imbalance
  • if feature importance is crutial
  • categorical features
  • if interpretebility is important
  • binary
  • if we want simple and interpreteble model
  • non-linear decision boundaries
  • If it is a credit risk analysis
  • data is high dimentional
  • local petterns matter more than global
  • large number of features and independece assumption
  • large dataset and complex relation
  • multiclass
  • multilabel
  • anomaly detection
  • outliears
  • if data is textual
  • if overfitting is a concern
  • if training time is a constraint
  • if data has hierarchical structure
  • if data distribution is unknown and changing
  • if real time prediction required

• nlp

  • EDA
    • token and text length distribution
    • vocabulary size
    • most frequent words
    • rare words
    • stop words and punctuations
    • N-grams analysis
    • Spelling and typos
    • Text annotation quality
    • entity co-occurance
    • error analysis
    • data augmentation oppurtunities
  • If Dimentionality reduction needed
  • If dealing with multilingual text
  • If handling noisy or informal text
  • text segmentation
  • chat bot
  • NER recognition
  • spam detection
  • POS tagging
  • transalation
  • Text summerisation
  • Speech recognition
  • Question answering
  • Sentiment analysis
    • In the case of training
    • using pretrained models
  • document similarity
  • Text generation
  • Coreference resolution
  • Dependancy parsing
  • Semantic role labeling
  • if real time processing required

• computer vision

  • if dealing with small datasets
  • if real time processing required
  • If there are many small objects in image
  • If needing high accuracy and precision
  • If handling noisy or varied environments
  • if real time object tracking is required
  • Data augmentation oppurtunities
  • Image classification
  • Object detection
  • Image segmentation
  • Instance segmentation
  • Image generation or style transfer
  • Image captioning
  • Image super-resolution
  • Medical image analysis
  • extract info from docs
  • Object tracking
  • Anomaly detection in image

• recommendataion system

  • If contains a lot of missing values
  • Addressing cold start problem
  • handling large scale data
  • If real time recommendation is important
  • If dealing with multicriteria recommendations
  • If incorporating social network information
  • colaborative filtering content based filtering
  • content based filtering
  • Matrix factorisation

• time series

  • Anomalies and outliers
  • If data shows Non-stationary
  • Handling missing values
  • Dealing with structural breaks
  • If forecasting future values
  • time series classification
  • seasonal adjustments
  • If work with long time series