DSST MIS cheat sheet

Bandwidth is the range of frequencies a channel carries; throughput is the actual data rate achieved, always lower than raw bandwidth due to protocol overhead and errors.

Nyquist theorem for a noiseless channel: max bit rate = 2 × bandwidth × log2(number of signal levels).

Shannon capacity: C = B × log2(1 + S/N), where S/N is the linear signal-to-noise ratio; 10 dB SNR = 10:1, 20 dB = 100:1.

OSI layers in order: Physical, Data Link, Network, Transport, Session, Presentation, Application; TCP/IP collapses these to Network Access, Internet, Transport, Application.

Circuit switching reserves a dedicated path for the call duration; packet switching shares links via statistical multiplexing and is more efficient for bursty data.

CSMA/CD detects collisions after transmission (Ethernet); CSMA/CA avoids them before transmission with ACKs and is required on wireless LANs.

Latency components: propagation (distance ÷ speed of light), transmission (bits ÷ link rate), processing, and queuing delay.

Subnetting: usable hosts per subnet = 2^h − 2, where h is the number of host bits; /24 leaves 8 host bits and 254 usable addresses.

Attenuation is signal loss over distance (dB), crosstalk is coupling between adjacent wires, and noise is any unwanted energy that degrades the signal.

CIA triad: confidentiality (encryption, access control), integrity (hashing, checksums), availability (redundancy, DDoS defense).

Symmetric ciphers (AES, 3DES) use one shared key and are fast; asymmetric ciphers (RSA, ECC) use key pairs and solve key distribution but are far slower.

Digital signature = hash of the message encrypted with the sender's private key; verifier uses the sender's public key, giving authentication, integrity, and nonrepudiation.

TLS handshake: negotiate cipher suite, authenticate server via certificate, exchange key material, then switch to symmetric encryption for bulk data.

Firewall types: packet-filtering (header fields only), stateful inspection (tracks connection state), and application proxy (inspects payload per protocol).

IDS detects and alerts; IPS detects and blocks inline. Signature-based detection matches known patterns; anomaly-based detection flags deviation from a baseline.

Phishing, spear phishing (targeted), and whitelisting/blacklisting are access-control policies; least privilege and separation of duties limit insider risk.

Risk = threat × vulnerability × impact; a vulnerability is a weakness, a threat is an actor or event that could exploit it, and exposure is the potential loss.

SDLC phases: planning, analysis, design, implementation, and maintenance; maintenance typically consumes the largest share of total lifetime cost.

Waterfall is sequential with frozen requirements; iterative/incremental and agile deliver working software in repeated short cycles with continuous feedback.

Data flow diagram symbols: process (circle), data store (open rectangle), external entity (square), data flow (arrow); a context diagram shows only the system and external entities.

DFD leveling rule: a parent process decomposes into a child diagram whose external data flows must match the parent's flows (balancing).

Entity-relationship model: entity, attribute, relationship; cardinality 1:1, 1:N, M:N, and M:N relationships require an associative table when implemented relationally.

Feasibility types: technical, economic (cost-benefit, ROI, payback), operational, schedule, and legal; economic feasibility is expressed in monetary terms.

Gantt charts show task duration and schedule overlap; PERT/CPM networks show dependencies and identify the critical path with zero slack.

Joint application design (JAD) gathers users and developers in structured workshops to define requirements faster than serial interviews.

Prototyping clarifies ambiguous requirements; a throwaway prototype is discarded, while an evolutionary prototype becomes the delivered system.

Decision types: structured (routine, rule-based, operational), semi-structured (some judgment), and unstructured (novel, strategic, high uncertainty).

Expected monetary value (EMV) = sum of (probability × payoff) for each alternative; choose the alternative with the highest EMV under risk.

Maximax picks the best possible outcome (optimistic), maximin picks the best worst case (pessimistic), and minimax regret minimizes the maximum opportunity loss.

Cost-benefit analysis: net present value discounts future cash flows to today; a positive NPV means the project adds value at the chosen discount rate.

Break-even quantity = fixed costs ÷ (unit price − unit variable cost); contribution margin per unit is price minus variable cost.

Payback period = time until cumulative cash inflows equal the initial investment; it ignores cash flows after payback and the time value of money.

A decision support system combines data, models, and a user interface to help managers with semi-structured problems; it supports, not replaces, judgment.

Groupthink and anchoring bias distort group decisions; Delphi technique collects anonymous expert forecasts over rounds to reduce dominance effects.

Explicit knowledge is codified and transferable (documents, databases, procedures); tacit knowledge is personal, experiential, and hard to articulate.

SECI model: Socialization (tacit to tacit), Externalization (tacit to explicit), Combination (explicit to explicit), Internalization (explicit to tacit).

Communities of practice are informal groups bound by shared expertise; they excel at transferring tacit knowledge where formal documents fail.

Knowledge management systems include repositories (best practices, lessons learned), expertise locators, and collaboration tools.

Intellectual capital components: human capital (skills), structural capital (processes, databases), and relational/customer capital (relationships, brand).

Data, information, knowledge hierarchy: data are raw facts, information is processed data with context, knowledge is actionable understanding applied to decisions.

Incentive design matters: reward knowledge sharing and reuse, because hoarding is rational when contribution is unrewarded and visible only as time cost.

Taxonomy and metadata enable retrieval; uncontrolled repositories decay into duplication and stale content without ownership and review cycles.

OLTP systems optimize for many small concurrent transactions; data warehouses optimize for large analytical queries and are read-mostly.

ETL: extract from sources, transform (clean, conform, aggregate), load into the warehouse; ELT loads raw data first and transforms inside the warehouse.

Star schema: a central fact table with foreign keys and measures, surrounded by denormalized dimension tables; snowflake schema normalizes dimensions into hierarchies.

Fact table grain defines what one row represents; declare grain before choosing dimensions or measures.

Measures are additive (quantity, revenue), semi-additive (balance, cannot sum across time), or non-additive (ratios, percentages).

Slowly changing dimensions: Type 1 overwrites (loses history), Type 2 adds a new row with version/effective dates (preserves history), Type 3 adds a prior-value column.

Data mart is a subset of a warehouse for one department; an operational data store holds near-real-time integrated data for operational reporting.

Conformed dimensions use identical keys and attributes across fact tables, enabling drill-across queries between business processes.

Data warehouse vs data lake: warehouses store structured, schema-on-write data; lakes store raw structured and unstructured data with schema-on-read.

CRISP-DM phases: business understanding, data understanding, data preparation, modeling, evaluation, deployment; the process is iterative, not linear.

Supervised learning uses labeled data (classification, regression); unsupervised learning finds structure without labels (clustering, association rules).

Association rule metrics: support = P(A and B), confidence = P(B|A), lift = confidence ÷ support(B); lift above 1 means positive correlation.

Apriori principle: if an itemset is frequent, all its subsets are frequent; if an itemset is infrequent, all supersets are infrequent - used to prune candidates.

Classification accuracy alone misleads on imbalanced data; use precision, recall, F1, and the confusion matrix (true/false positives and negatives).

k-means partitions into k clusters by minimizing within-cluster distance; k must be chosen in advance and results depend on initial centroids.

Overfitting means the model fits training noise and generalizes poorly; detect with holdout validation or cross-validation, mitigate with pruning or regularization.

Decision tree splits are chosen by information gain (entropy reduction) or Gini impurity reduction; deep trees overfit unless pruned.

Regression predicts a continuous value; classification predicts a discrete class. Correlation between variables does not establish causation.